Agents: The Docs Agent and the Marketing Agent: which setting to use to give them context, starting work from the editor, GitHub, GitLab, Slack, Discord, Jira or an MCP client, Docs Agent and Marketing Agent tasks, Automations and automatic pull request review, and how EkLine opens pull requests # Agents overview > What the Docs Agent and the Marketing Agent write, where their work goes, how to start each one, and who can use them. EkLine has two agents: the Docs Agent updates your documentation, and the Marketing Agent drafts blog posts, LinkedIn posts and newsletters. ## See the Docs Agent in action [Section titled “See the Docs Agent in action”](#see-the-docs-agent-in-action) [How to automate documentation workflows with Slack and EkLine](https://www.youtube.com/embed/xn3cq_dfQVY) ## What the Docs Agent and Marketing Agent do [Section titled “What the Docs Agent and Marketing Agent do”](#what-the-docs-agent-and-marketing-agent-do) | | Docs Agent | Marketing Agent | | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------- | | **What it writes** | New and updated docs pages, such as guides, API references, READMEs and changelogs. It also answers questions about your docs. | Blog posts, LinkedIn posts and newsletters, in your company’s voice. | | **Where its work goes** | A pull request on GitHub, or a merge request on GitLab, in a repository you listed under **Where EkLine writes**. | A draft in the editor that you copy or download. | | **Who approves it** | Your team reviews and merges the pull request. | You publish the draft yourself. | ### Docs Agent [Section titled “Docs Agent”](#docs-agent) The Docs Agent drafts its changes in the editor, where you accept, reject or edit them. When the draft is ready, click **Raise PR**, which puts “Open a pull request” in the message box, and send it. Sessions started from a pull request comment, Slack or an automation can open a pull request without that step. The agent writes only to the repositories under **Where EkLine writes**. It reads those under **What else EkLine can read** but does not write to them. See [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/). The Docs Agent can also draft changes to a Confluence or Pylon knowledge base, which you publish with **Update KB**. See [Manage a knowledge base](/agent/manage-knowledge-base/). ### Marketing Agent [Section titled “Marketing Agent”](#marketing-agent) Each Marketing Agent draft opens in the editor as a document you can edit. **Copy** copies the open draft. **Download** saves the session’s drafts as Markdown, or as HTML from the menu. The agent opens a pull request only when you ask it to publish a draft to one of your documentation repositories and say where. It does not publish to your website or CMS. See the [Marketing Agent overview](/marketing-agent/). ## Start a session [Section titled “Start a session”](#start-a-session) **Docs Agent** * **In the app:** click **New session** in the sidebar, or enter a prompt in **Start a conversation** on the dashboard. If the composer shows **Docs** and **Marketing**, select **Docs**. See [Work in the editor](/agent/editor/). * **From a pull request or merge request:** mention `@ekline-ai` in a comment. See [GitHub pull requests](/agent/github-integration/) and [GitLab merge requests](/agent/gitlab-integration/). * **From Slack or Discord:** mention EkLine in a channel or thread. See [Slack](/agent/slack-bot/) and [Discord](/agent/discord-bot/). * **From Jira:** a ticket that matches one of your triggers starts a session. See [Jira tickets](/agent/jira-triggers/). * **On a schedule:** see [Automations](/agent/scheduled-agents/). * **From a Journeys report:** click **Fix in docs** on a finding. See [Turn a gap into a fix](/journeys/turn-a-gap-into-a-fix/). * **From an MCP client** such as Claude Code or Cursor. See [MCP clients](/agent/mcp/). **Marketing Agent** * **In the app:** click **New session** in the sidebar, or enter a prompt in **Start a conversation** on the dashboard. If the composer shows **Docs** and **Marketing**, select **Marketing**. See [Start a session](/marketing-agent/#start-a-session). * **From a Journeys report:** click **Draft in Marketing** on a finding, or **Start the fix** on a collection report. See [Write a post that answers a buyer question](/marketing-agent/buyer-question-post/). * **From an MCP client.** See [MCP clients](/agent/mcp/). Slack, Discord, Jira and pull request comments start Docs Agent sessions only. An [automation](/agent/scheduled-agents/) starts a session of the agent it was created for, the Docs Agent or the Marketing Agent. ## Who can use the agents [Section titled “Who can use the agents”](#who-can-use-the-agents) Any member of your organization can start a session with either agent. Both agents are available to every organization, including free accounts. If you don’t see Automations, Jira triggers, knowledge base management or a settings tab these pages describe, contact . See [Credits and usage](/workspace/credits-and-usage/) for how agent work is counted. ## What is in this section [Section titled “What is in this section”](#what-is-in-this-section) | Part | Pages | | --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Give agents context | [Which setting to use](/agent/which-setting-to-use/), [Company Profile and Domains](/agent/company-profile/), [Custom instructions](/agent/custom-instructions/), [Reference documents and repository notes](/agent/reference-documents/), [Skills](/agent/custom-skills/) | | Start work | [Work in the editor](/agent/editor/), [Find a past session](/agent/find-sessions/), [GitHub pull requests](/agent/github-integration/), [GitLab merge requests](/agent/gitlab-integration/), [Slack](/agent/slack-bot/), [Discord](/agent/discord-bot/), [Jira tickets](/agent/jira-triggers/), [MCP clients](/agent/mcp/), [Sandbox](/agent/sandbox/) | | Docs Agent tasks | [Create documentation](/agent/create/), [Keep docs in sync with code](/agent/keep-docs-in-sync-with-code/), [Generate release notes](/agent/release-notes/), [Manage a knowledge base](/agent/manage-knowledge-base/), and the other pages under **Docs Agent tasks** | | Marketing Agent tasks | [Marketing Agent overview](/marketing-agent/), [Write a post that answers a buyer question](/marketing-agent/buyer-question-post/), [Turn shipped changes into launch content](/marketing-agent/launch-content/), [How claims stay accurate](/marketing-agent/accurate-claims/) | | Automate | [Automations](/agent/scheduled-agents/), [Automatic PR review](/agent/automatic-pr-review/), [Control which PRs are reviewed](/agent/review-rules/) | | Review and ship | [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/), [For engineers: an EkLine pull request or comment on your repository](/agent/for-engineers/) | ## FAQ [Section titled “FAQ”](#faq) ### Which agent should I use? [Section titled “Which agent should I use?”](#which-agent-should-i-use) The Docs Agent for your documentation, or a Confluence or Pylon knowledge base. The Marketing Agent for a blog post, a LinkedIn post or a newsletter. ### Do the two agents share settings? [Section titled “Do the two agents share settings?”](#do-the-two-agents-share-settings) Partly. Each agent has its own custom instructions and never reads the other’s. Both read your repository notes and use your rules and skills. Both can read your Company Profile and reference documents. See [Which setting to use](/agent/which-setting-to-use/). ### Where do I see what an agent did? [Section titled “Where do I see what an agent did?”](#where-do-i-see-what-an-agent-did) Sessions are listed under **Sessions** in the sidebar, grouped by date. Open one to see the agent’s messages and the changes it made. To search or filter them, see [Find a past session](/agent/find-sessions/). # Marketing Agent > What the Marketing Agent writes, which sources it reads, where its drafts go, and how to start a session. The Marketing Agent drafts blog posts, LinkedIn posts, and newsletters in your company’s voice from your own docs, code, pull requests, and tickets. It also writes outreach lists for getting your company named on the roundups, review sites and directories AI assistants cite. ## What it writes [Section titled “What it writes”](#what-it-writes) | Format | What you get | | ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | **Blog post** | A title, subheadings, images where they help, and a closing call to action. | | **LinkedIn post** | Paragraphs only, no headings, under 1,300 characters unless you ask for longer. The agent gives you any link in the chat, to post as a comment. | | **Newsletter** | A subject line, a two-sentence intro, two to five sections, and a sign-off, or a single section when you ask for one. | | **Outreach list** | The pages on other sites to get your company named on, ranked, each with a published contact, what to ask for, a short pitch, and an entry to paste. | For a blog post, the agent picks the type that fits the reader’s question, each with its own structure and length: explainer, decision guide or comparison (including “alternatives to” posts), implementation guide, migration guide, benchmark or audit, playbook, cost or ROI guide, point of view, or launch. See [Turn shipped changes into launch content](/marketing-agent/launch-content/). Images are illustrations and diagrams the agent draws, never photographs, at most three per draft. They use the colors in **Visual Style Guidelines** in your Company Profile, or a neutral palette when none are set. They cannot include logos, uploaded images, or named fonts. You can ask for a different image, a change, or none. ## What it reads [Section titled “What it reads”](#what-it-reads) Before it drafts, the agent reads your **Company Profile** (**Settings** > **Organization** > **Company Profile**): your product, audiences, competitors, brand names, **Tone and Voice**, and **Visual Style Guidelines**. If no tone and voice is on file, it still does the work, says so, and offers to help you write voice guidance and save it to your Company Profile. In an outreach list, it marks each pitch as written without your voice. See [Company Profile and Domains](/agent/company-profile/). It follows its own custom instructions, set in **Settings** > **Marketing Agent** > **Custom instructions**, which never apply to the Docs Agent. See [Custom instructions](/agent/custom-instructions/). Where you have connected them, it also reads: * Your documentation and code repositories, through GitHub or GitLab, including merged pull requests, changelog files, and release tags. * Closed tickets in Jira or Linear. * Knowledge bases and uploaded reference documents. See [Reference documents and repository notes](/agent/reference-documents/). * Launch documents and plans in Notion, Google Drive, Confluence, or Slack. * The links, files, and pull request numbers you add to your request, and a Journeys report when you start from one. * Public pages: your website and blog, competitors’ pages, and discussions on Reddit, Hacker News, and GitHub. Research is read-only, and a draft never links to or names a private document. See [How claims stay accurate](/marketing-agent/accurate-claims/). ## Where drafts go [Section titled “Where drafts go”](#where-drafts-go) Each draft opens in the Marketing Agent editor, where you can edit it directly. From the toolbar: * **Copy** copies the open draft. * **Download** saves every draft in the session, as Markdown by default, with HTML in the menu. Several drafts, or drafts with images, download as one ZIP file. The agent publishes only when you ask it to and name the destination. Ask it to add a draft to one of your documentation repositories and it confirms the path if you did not name one, then opens a pull request for your team to review and merge. The agent never overwrites an existing page. ## Start a session [Section titled “Start a session”](#start-a-session) 1. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Marketing**. The page reads **What are we making?** 2. Say what you want, for example `Write a blog post about our new audit log`. Or click **Write a blog post**, **Write a LinkedIn post**, or **Write a newsletter section** under the composer: it puts its opening words in the composer for you to finish. 3. Add anything the agent should use: paste links or pull request numbers, or click the attachment icon to add files. To approve the outline before it writes, add `Show me an outline first and wait for my go-ahead.` To get more than one format, ask for each one, for example `Write a blog post and a LinkedIn post about our new audit log`. 4. Send the message. The agent asks in the chat for anything else it needs. You can also start from the dashboard’s **Start a conversation** box (select **Marketing** when both agents are on), from a Journeys report (see [Write a post that answers a buyer question](/marketing-agent/buyer-question-post/)), or by typing `/blog-post`, `/linkedin-post`, or `/newsletter` and a topic in the chat. Once a draft is open, the shortcuts **Brand voice check**, **Claims check**, **Shorter**, and **Variant for LinkedIn** sit under the message box. Each adds a request for you to send. ## Access [Section titled “Access”](#access) Any member of your organization can start a Marketing Agent session. See the [Agents overview](/agent/) for both agents, and [Credits and usage](/workspace/credits-and-usage/) for how agent work is counted. ## What it does not do [Section titled “What it does not do”](#what-it-does-not-do) * **It does not publish to Webflow or any other CMS.** You publish the post on your site yourself, from a copy or a download. * **It does not start from Slack, Jira or pull request comments.** Those start Docs Agent sessions only. A Marketing Agent session starts from the editor, an [MCP client](/agent/mcp/), a Journeys report, or an [automation](/agent/scheduled-agents/) that runs as the Marketing Agent. * **It does not post or send for you.** You post the LinkedIn draft, send the newsletter, and send each outreach pitch yourself. ## FAQ [Section titled “FAQ”](#faq) ### Can I schedule the Marketing Agent? [Section titled “Can I schedule the Marketing Agent?”](#can-i-schedule-the-marketing-agent) Yes. Create a custom automation and select **Marketing Agent** under **Runs as**, or turn on **Release Marketing Brief**, which by default every Monday drafts a Slack-ready brief of the customer-facing features that shipped in the last 7 days, as raw material for your marketing team. A Marketing Agent automation does not open pull requests; its drafts are in the run’s session. See [Automations](/agent/scheduled-agents/). ### Does it use the Docs Agent’s custom instructions? [Section titled “Does it use the Docs Agent’s custom instructions?”](#does-it-use-the-docs-agents-custom-instructions) No. Each agent has its own. The Marketing Agent never reads text you save for the Docs Agent. ### Can one request produce more than one format? [Section titled “Can one request produce more than one format?”](#can-one-request-produce-more-than-one-format) Yes. Ask for each format in the same request, for example `Write a blog post and a LinkedIn post about our new audit log`. Each one is drafted as its own file. # How claims stay accurate > How the Marketing Agent ties each claim to a source, checks blog posts and drafts that name your product, and what you review before anything ships. The Marketing Agent ties each claim in a draft to a source it read in that session, and checks every blog post and every draft that names your product in a separate pass before you see it. These checks make an unsupported claim less likely, but your review is still the last check. ## Drafts start from your sources [Section titled “Drafts start from your sources”](#drafts-start-from-your-sources) The agent researches before it writes, and it writes from what it read in that session: * **Every claim about your product cites a source the agent read.** A claim with no source is cut, not softened. The agent never invents a feature, a number, a quote, or a customer name, and never says that something shipped, is generally available, or is not behind a flag unless a source says so. * **Evidence is labeled.** Each finding is an official statement, a measured result, a practitioner’s report, or the agent’s own inference, which is never written as a measured result. Numbers, quotes, and code keep the source’s exact words, and a search-result snippet is never evidence. * **Your own pages are compared.** For a price, plan, limit, integration, security, or availability fact about your product, research checks your published pages and flags the one to fix if two of them disagree. * **Links and code are traceable.** A blog post links only to sources the agent read, inline on the words each one supports. Code and commands appear only when copied from a file the agent opened. * **Private sources stay private.** A draft never links to or names a private document. It describes one in words a reader could find, such as “the release notes”. In the chat, the agent cites it by path or link. Without research findings, the agent stops and asks for sources. It does not draft a blog post from memory. For the sources it reads, see [Marketing Agent](/marketing-agent/). ## A separate claims check [Section titled “A separate claims check”](#a-separate-claims-check) After drafting, every blog post and every draft that names your product goes through a claims check. The check reopens each source it can and looks at each image. A claim is anything that says your product does, has, ships, supports, or measures something; any number; any named customer, company, or person; any quote; any date or status (beta, generally available, behind a flag); and any comparison with a competitor or with how people work today. Each claim gets one of four outcomes: | Outcome | Meaning | What happens to it | | -------------- | ------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------- | | `sourced` | A source states it, with the same scope. | It stays. | | `unsourced` | Nothing the check can read states it. | It is cut, or rewritten without the number. | | `contradicted` | A source says otherwise. | It is rewritten to match the source, or cut. | | `overstated` | A source supports a narrower claim: for example, the draft drops a date, a sample, or a plan, or states one result as general. | It is narrowed to what the source shows. | The check also flags a link to a page outside the draft’s sources, and a diagram that differs from the text. Everything the check flags is fixed before you see the draft. For a blog post, if the fixes changed the draft, the check runs again on the final version and anything still open is reported to you. A blog post also gets a separate voice review, whose fixes keep each fact’s sourced wording. To run the check again after you edit, click **Claims check** under the message box and send the message. ## What you review before anything ships [Section titled “What you review before anything ships”](#what-you-review-before-anything-ships) 1. **The chat summary.** It opens with the agent’s recommendation, then lists what it read, with sources, what it could not read, what it drafted, which images it added, any code or commands in the draft it did not run, and the assumptions it made. 2. **The draft**, open in the editor, where you can edit it directly. When a later turn changes a draft, the editor shows the changes with **Accept All** and **Reject All**. 3. **The publishing step.** The agent publishes nothing unless you ask and name the destination; otherwise you copy or download the draft and publish it yourself. A draft added to a documentation repository arrives as a pull request your team reviews and merges. ## FAQ [Section titled “FAQ”](#faq) ### Can a draft still contain a wrong claim? [Section titled “Can a draft still contain a wrong claim?”](#can-a-draft-still-contain-a-wrong-claim) Yes. The check compares the draft with the sources the agent could read, and a source can be out of date or wrong. Read the summary’s list of what the agent could not read, and check any claim that rests on one of those. ### Does the agent make up a customer quote or a statistic? [Section titled “Does the agent make up a customer quote or a statistic?”](#does-the-agent-make-up-a-customer-quote-or-a-statistic) It should not. A number, quote, or customer name needs a source the agent read, and the claims check marks one without a source as `unsourced`, which is then cut or rewritten. Before you publish, check each quote and statistic against the sources in the chat summary. ### Does a LinkedIn post or newsletter get checked too? [Section titled “Does a LinkedIn post or newsletter get checked too?”](#does-a-linkedin-post-or-newsletter-get-checked-too) Yes, when it names your product. A draft that does not name your product skips the check. # Write a post that answers a buyer question > Have the Marketing Agent draft a blog post that answers one question your buyers ask, from a Journeys report or from your own question. The Marketing Agent can draft a blog post that answers one question your buyers ask. Start from a gap a Journeys report found, or from a question you already have. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * Any member of your organization can start a Marketing Agent session. * Your Company Profile has **Tone and Voice** filled in, under **Settings** > **Organization** > **Company Profile**. Without it, the agent asks about your audience, tone, and product before it drafts. See [Company Profile and Domains](/agent/company-profile/). * To start from a report: Journeys and a report that found a gap. See [Read a Buyer questions report](/journeys/read-a-report/). * Optional: your documentation repository connected, so the agent can read the pages the post should point to. ## Steps [Section titled “Steps”](#steps) * From a Journeys report 1. Open the report and find a fix for the Marketing Agent: * In a Buyer questions report, go to **Fix these first** and find a card that reads **The Marketing Agent can draft this**. Click **Start the fix**. You stay on the report, and the button changes to **View in Marketing Agent**. * In a single journey’s report, go to **What went wrong** and find a cause with **Draft in Marketing**. Click it, and the app opens the new session. The report suggests which agent fits each fix. See [Turn a gap into a fix](/journeys/turn-a-gap-into-a-fix/). 2. Open the session. The agent reads the report’s evidence, including the answers AI assistants gave, and checks whether your existing content already covers the gap. If a page already covers the gap, the agent says so or drafts changes to that page, instead of a duplicate. If not, it drafts what it thinks fits and tells you why. * From your own question 1. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Marketing**. 2. Type `Write a blog post that answers:` and the question, the way a buyer would ask it. 3. Add the audience and the goal if you know them. If you leave one out, the agent fills it in from your Company Profile and lists it in the chat as an assumption. It asks only when your profile cannot settle it and the answer would change the post. 4. Paste the links, docs paths, or pull request numbers the answer depends on. To add files, click the attachment icon. 5. Optional: add `Show me an outline first and wait for my go-ahead.` 6. Send the message. Either way, finish the same way: 1. Answer any questions the agent asks. It asks only when the answer would change the post. 2. Read the summary the agent posts in the chat: what it read, with sources, what it could not read, and the assumptions it made. 3. Edit the draft in the editor. To recheck it after you edit, click **Claims check** or **Brand voice check** under the message box and send the message. 4. Click **Copy**, or **Download** the post as Markdown or HTML, and publish it on your blog yourself. To add it to one of your documentation repositories instead, see [Where drafts go](/marketing-agent/#where-drafts-go). ### Example: Acme [Section titled “Example: Acme”](#example-acme) Acme sells a payments API. Its Buyer questions report shows that when buyers ask “Which payments API handles split payouts for marketplaces?”, AI assistants recommend Hooli. Acme’s request: ```text Write a blog post that answers: Which payments API handles split payouts for marketplaces? Audience: engineering leads at marketplace companies. Goal: get readers to Acme's split payouts guide. Source: https://docs.acme.com/payouts/split ``` The agent searches the question the way a buyer would, reads the top results, and notes what they leave unanswered. If a post on Acme’s blog already answers it fully, the agent asks whether to write a new post or refresh that one. The draft answers the question within its first three paragraphs, and adds an FAQ only when the agent found three or more real questions readers ask about it. ## Check it worked [Section titled “Check it worked”](#check-it-worked) * The draft is open in the editor. A blog post’s title is the buyer’s question or its answer. * The chat summary lists the sources the post rests on. * If you started from a report, the card or cause now shows **View in Marketing Agent**. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | What you see | What to do | | ----------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- | | The cause shows **Fix in docs**, not **Draft in Marketing** | The report suggests a docs fix for this cause. Use the Docs Agent, or start a Marketing Agent session with the question yourself. | | “Marketing Agent did not start a session. Try again.” | Click the button again. | ## Next [Section titled “Next”](#next) * [How claims stay accurate](/marketing-agent/accurate-claims/): what the agent checks before you see a draft. * [Track results over time](/journeys/track-results/): after you publish, run the journey again to see whether the answers change. * [Turn shipped changes into launch content](/marketing-agent/launch-content/) # Turn shipped changes into launch content > Have the Marketing Agent turn merged pull requests, a changelog, or a release into a launch post, a LinkedIn post, or a newsletter draft. The Marketing Agent can turn what your team shipped (merged pull requests, changelog entries, release tags, and closed tickets) into a launch post, a LinkedIn post, or a newsletter draft. Nothing starts on its own when you merge or release: you start the session. For the whole job, with docs and release notes from the Docs Agent, follow the recipe [Turn a release into docs, release notes and a launch post](/get-started/recipes/release-to-launch/). ## Before you begin [Section titled “Before you begin”](#before-you-begin) * Any member of your organization can start a Marketing Agent session. * The repositories that hold your code and changelog are connected through [GitHub](/agent/github-app-setup/) or [GitLab](/agent/gitlab-setup/). * To include closed tickets, Jira or Linear is connected. See [Integrations](/agent/integrations/). * Your Company Profile has **Tone and Voice** filled in, under **Settings** > **Organization** > **Company Profile**. ## Steps [Section titled “Steps”](#steps) 1. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Marketing**. 2. Ask for a format and say what shipped, for example `Write a blog post about the releases since 1 September`. **Write a blog post**, **Write a LinkedIn post**, and **Write a newsletter section** under the composer put the opening words in for you. Either give a time window, or name the changes: * A window, for example “releases since 1 September”. Without one, the agent looks back 14 days. The longest window is 90 days. * Specific changes: paste pull request numbers, a release tag, or a changelog path into your request. 3. To get more than one format from the same changes, ask for each one in the same request, for example `Write a blog post and a LinkedIn post about the releases since 1 September`. 4. Send the request, then answer the agent’s questions. For a launch post, it asks when no source says who can use the feature today (plan, version, region, or beta), or where the call to action should point. 5. Read the agent’s sizing for a launch post. It recommends one of: a changelog line and no post, a product update post (600 to 1,000 words), a launch article (1,200 to 2,000 words), or a company launch that also needs a launch plan. When it recommends a changelog line, it gives you the line in the chat and still writes the post you asked for. 6. Review the draft in the editor, then click **Copy**, or **Download** it as Markdown or HTML, and publish it yourself. To add it to one of your documentation repositories instead, see [Where drafts go](/marketing-agent/#where-drafts-go). ## What the agent includes and leaves out [Section titled “What the agent includes and leaves out”](#what-the-agent-includes-and-leaves-out) **A launch post says exactly who can use the feature today**, by plan, version or release, region, or beta. A feature behind a flag, in beta, or on the roadmap is never described as available. The post also covers what changes for the reader (a before and after), objections taken only from your sources such as support threads, sales notes, or documented limits, and one call to action whose link the agent opens to check it works. **A release roundup leaves out** any change that was: * reverted in the window * a security or vulnerability fix * behind a feature flag, with no sign it is fully rolled out * internal only: refactors, infrastructure, build tooling, tests, or dependency updates with no visible effect for users * a documentation-only change It also drops a change when no source explains why it matters to users. A newsletter has two to five sections, one per change, ordered by what matters most to its audience. If you ask for a single newsletter section, it covers the top changes in one section. If every change is excluded, that is a quiet window, not an error. Release notes and changelog pages are Docs Agent work, not Marketing Agent formats. See [Generate release notes](/agent/release-notes/). ## Check it worked [Section titled “Check it worked”](#check-it-worked) * The draft is open in the editor. * The chat summary lists the pull requests, tickets, or changelog entries the draft rests on, with links. * A launch post includes a line that says who can use the feature today. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | What you see | What to do | | ----------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The agent says it cannot reach any source | Connect the repository through GitHub or GitLab, or paste the pull requests or changelog into your request. | | The agent asks for a shorter window | It works on at most 90 days at a time. Split the period into smaller windows. | | A change you expected is missing | It matched one of the exclusions above, or the agent judged that users would not notice it. If it was behind a flag and is now fully rolled out, point the agent to the release note, pull request, or ticket comment that says so. | ## Next [Section titled “Next”](#next) * [Turn a release into docs, release notes and a launch post](/get-started/recipes/release-to-launch/) * [How claims stay accurate](/marketing-agent/accurate-claims/) * [Write a post that answers a buyer question](/marketing-agent/buyer-question-post/) # Automatic documentation review on pull requests > EkLine automatically detects when pull requests need documentation updates and suggests or creates them based on confidence level. When you open a pull request, EkLine checks whether your docs need an update. It rates how sure it is. Then it stays silent, posts a suggestion, or starts a docs update for you. ## How it works [Section titled “How it works”](#how-it-works) ``` flowchart TD A[Pull request opened] --> B[EkLine analyzes PR diff] B --> C{Documentation impact?} C -->|Low confidence| D[No action taken] C -->|Medium confidence| E[Posts suggestion comment] C -->|High confidence| F[Posts auto-trigger comment] F --> G[Docs update flow runs automatically] G --> H[Documentation PR created] E --> I[You decide whether to trigger] ``` EkLine checks a pull request when it is opened, marked ready for review, or has a label added or removed. It: 1. Reads the PR diff, title, and description. 2. Compares the changes with your docs. 3. Decides whether the changes affect your docs. 4. Acts on how sure it is. Draft pull requests are skipped unless you turn off **Skip draft PRs** in the repository’s [review rules](/agent/review-rules/). ## Confidence tiers [Section titled “Confidence tiers”](#confidence-tiers) EkLine puts each check into one of three tiers: | Tier | When it applies | What EkLine does | | ---------- | ---------------------------------------------------------------------------------------- | ------------------------------------------- | | **Low** | Refactoring, new tests, or bug fixes that users do not see | Stays silent. Posts no comment. | | **Medium** | Changes in behavior that might affect a documented feature | Posts a suggestion comment on the PR | | **High** | New API endpoints, renamed configuration, removed features, or changed public interfaces | Posts a comment that starts the docs update | EkLine starts a docs update only when the impact is clear. When it is unsure, it stays silent. ### Medium confidence [Section titled “Medium confidence”](#medium-confidence) EkLine posts a comment on the pull request that says which docs might need an update. The comment mentions `@ekline-ai`, so you can start the update yourself if you agree. EkLine does nothing more. You decide whether to go ahead. ### High confidence [Section titled “High confidence”](#high-confidence) EkLine posts a comment that starts the [GitHub PR bot](/agent/github-integration/) flow. EkLine replies that the Docs Agent is on it, with a link to the session. The agent drafts the docs update and opens a docs pull request. If **Auto-create documentation PRs** is off in the repository’s review rules, the drafts wait in the editor instead. If no pull request appears, open the session from the link and ask the agent to open one. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) Before automatic PR review works on your repositories, you need: * An EkLine organization account. * The [EkLine GitHub App](https://ekline.io/settings) installed on the repository. * Automatic PR review on for your organization. It is on by default. * The repository **selected in your GitHub integration settings**, under either **Where EkLine writes** or **What else EkLine can read**. EkLine reviews pull requests only from repositories you selected. It skips PRs from other repositories the app can access. See [Configure repositories](/agent/github-app-setup#configure-repositories). * **Enable monitoring** on for the repository. It is on by default. Then it runs on the repositories you selected, for pull requests that pass each repository’s [review rules](/agent/review-rules/). ## What you see on the pull request [Section titled “What you see on the pull request”](#what-you-see-on-the-pull-request) | Scenario | What appears on the PR | | ---------------------- | --------------------------------------------------------------------------------------------------------------- | | Low confidence | Nothing. EkLine stays silent. | | Medium confidence | A comment that suggests a docs update, with `@ekline-ai` mentioned so you can start it | | High confidence | A comment that starts the docs update, then a reply with a link to the session. The agent then opens a docs PR. | | Error during the check | Nothing. EkLine stays silent. | At most one automatic review comment appears per pull request. ## Relationship to the GitHub PR bot [Section titled “Relationship to the GitHub PR bot”](#relationship-to-the-github-pr-bot) Automatic PR review builds on the [GitHub PR bot](/agent/github-integration/): * **Manual flow**: You mention `@ekline-ai` in a PR comment to start a docs session. * **Automatic flow**: EkLine posts the `@ekline-ai` mention for you when the impact is clear. Both flows start the same kind of docs session. If EkLine already posted an automatic comment and you then mention `@ekline-ai` on the same PR, the existing session resumes. No second session starts. ## GitLab merge requests [Section titled “GitLab merge requests”](#gitlab-merge-requests) Automatic review also runs on GitLab merge requests, with these differences: * A merge request is reviewed when it is opened, reopened, or marked ready. Adding or removing a label, or pushing new commits, does not start a review. * EkLine needs the webhook and the access token from [Connect GitLab](/agent/gitlab-setup/). * Merge requests from `dependabot`, `renovate`, `gitlab-bot`, or any username ending in `-bot` are always skipped. * EkLine’s comments are posted as the GitLab user who owns the access token you connected. ## Run a manual assessment [Section titled “Run a manual assessment”](#run-a-manual-assessment) You can also check a PR yourself with the `/analyze-pr-for-docs-needs` command in the EkLine Docs Agent. Give the PR URL as an argument: ```text /analyze-pr-for-docs-needs https://github.com/your-org/your-repo/pull/123 ``` The agent compares the PR diff with your docs and uses the same confidence tiers. Use it to check a PR without waiting for the automatic review, or to check a PR again after large changes. ## Current limitations [Section titled “Current limitations”](#current-limitations) * **One check per PR**: New commits pushed after the first check do not start a new one. * **Fixed tiers**: You cannot change the thresholds that decide whether EkLine stays silent, suggests, or starts an update. To control which pull requests EkLine reviews, use [review rules](/agent/review-rules/). * **PR context only**: The check uses the PR diff, title, and description. It does not read linked Jira or Linear tickets. * **This PR’s changes only**: EkLine checks the changes in the PR. It does not look for gaps that were already in your docs. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | No comment appears on my PR | Check that the repository is selected in your GitHub integration settings, under **Where EkLine writes** or **What else EkLine can read**. Check that **Enable monitoring** is on for it. Check that the EkLine GitHub App is installed and automatic PR review is on for your organization. EkLine also skips draft PRs by default, PRs your review rules filter out, and PRs that already have a session. A Low check posts no comment. | | EkLine posted a suggestion but I want the update | Mention `@ekline-ai` in a new comment on the PR. Say which docs changes you need. | | The auto-triggered docs PR is wrong | Comment on the docs PR, mention `@ekline-ai`, and give your feedback. The agent resumes the session and pushes new commits. | | EkLine started an update on a PR that does not need docs | The check is best-effort. Close or ignore the docs PR. | ## Next steps [Section titled “Next steps”](#next-steps) * [Control which PRs are reviewed](/agent/review-rules/): scope automatic PR review per repository by branch, title, author, label, or changed files. * [GitHub PR bot](/agent/github-integration/): trigger documentation sessions manually from PR comments. * [Create documentation](/agent/create/): generate documentation from your codebase, videos, and external sources. * [Update and review](/agent/update-review/): keep documentation in sync with code changes. # Combine multiple sources into one document > Reference a ticket, pull request, Slack thread, and spec in a single Docs Agent prompt to produce one authoritative document, then open a pull request. Give Docs Agent several sources in one prompt, and it combines them into one document. You review it and open it as a pull request. Combine sources when a feature’s details are spread across tools: requirements in a Jira or Linear ticket, code in a pull request, edge cases in a Slack thread, and the design in a Notion or Confluence spec. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * At least one documentation repository [connected to EkLine](/agent/github-app-setup/). * The integrations that hold your sources connected under **Settings** > **Integrations**. See [Integrations](/agent/integrations/) to connect Slack, Notion, Linear, Jira, Confluence, Google Drive, GitLab, and PostHog. * Access to each source for the account that connected its integration. Integrations are connected once for your whole organization, and the agent reads through that connection, not with your own access. ## Combine your sources [Section titled “Combine your sources”](#combine-your-sources) 1. Gather the sources that describe the same topic. For a feature launch, that is often the ticket, the pull request, the Slack thread where the team discussed it, and the original spec. 2. Open Docs Agent and write one prompt that lists every source and says what to produce. Paste exact URLs or ticket IDs. The agent works out which tool each one comes from: ```text Write a feature guide for our new rate limiting, drawing on: - Linear ticket ENG-1234 for the requirements - GitHub pull request #482 for the implementation details - This Slack thread for the edge cases the team found: https://workspace.slack.com/archives/C01234/p1234567890 - The original design in Notion: https://notion.so/your-workspace/rate-limiting-spec Put it in the docs repo under guides/. Cover what rate limiting does, the default limits, the response headers, and how to handle a 429 error. ``` 3. Wait while the agent fetches each source and drafts the document. It reports the sources it read as it works. 4. Review the draft in the editor panel. Confirm that each source contributed the part you expected: requirements from the ticket, behavior from the pull request, edge cases from the thread. ## Write an effective multi-source prompt [Section titled “Write an effective multi-source prompt”](#write-an-effective-multi-source-prompt) Give each source a purpose and name the output you want. | Instead of… | Try… | | ------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | | “Document this feature from ENG-1234 and PR #482” | “Write a feature guide for rate limiting. Use ENG-1234 for the requirements and PR #482 for the implementation details.” | | “Combine these links into docs” | “Create one deployment guide from the runbook in Confluence and the checklist in this Slack thread. Prefer the runbook where they disagree.” | | “Update the API docs with everything about auth” | “Update the authentication reference using ENG-1234 (the OAuth change) and PR #501 (the new token endpoint). Keep the existing structure.” | **Include in your prompt:** * Each source as an exact URL or ticket ID. * What each source contributes. * The document type and where it belongs in your repository. * The sections you want, so the agent maps sources onto structure. ## Handle conflicting sources [Section titled “Handle conflicting sources”](#handle-conflicting-sources) A ticket describes the plan, but the pull request may ship something different, and a spec may be older than the final design. When sources disagree, tell the agent which one wins: ```text Where the Notion spec and PR #482 disagree, follow the pull request, because it reflects what actually shipped. Note any requirement from ENG-1234 that the PR did not implement. ``` If you don’t say which wins, the agent points out the conflict in the chat so you can decide. Ask it to settle a specific difference, and it revises that section. ## Verify before publishing [Section titled “Verify before publishing”](#verify-before-publishing) 1. Enable **View All Changes** in the toolbar to see a diff of everything the agent wrote. Confirm no source was dropped and no section mixes up details from different sources. 2. Check that claims trace back to a source. If a detail looks unsupported, ask the agent where it came from, for example `Which source says the default limit is 100 requests per minute?` 3. Click **Raise PR** in the toolbar. The chat box fills with `Open a pull request for` and your repository’s name. Press **Enter** to send it, or edit it first to add instructions. 4. The agent opens the pull request and replies with a link where you finish the review on GitHub. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause | Fix | | ------------------------------------- | ---------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | A source was skipped | The integration isn’t connected, or the account that connected it can’t open that source | Connect the integration under **Settings** > **Integrations**, or give the connected account access to the source, then reference it again. | | The draft leans on one source | The prompt didn’t say what each source contributes | Restate the prompt and assign a role to each source, such as “use the ticket for requirements, the PR for behavior.” | | Details from two sources are mixed up | The sources overlap and no precedence was set | Tell the agent which source wins for the overlapping topic, then ask it to revise that section. | | The agent asks which source to trust | Two sources give conflicting facts | Reply with the source of truth, for example “follow the pull request; the spec is outdated.” | | A pasted link isn’t recognized | The URL is malformed, or the service isn’t connected | Paste the exact URL from the source, and confirm the matching integration is connected. | ## Next steps [Section titled “Next steps”](#next-steps) * [Integrations](/agent/integrations/): connect Slack, Notion, Linear, Jira, Confluence, Google Drive, and more. * [Create documentation](/agent/create/): generate READMEs, API references, and guides from a single source. * [Update and review](/agent/update-review/): keep documentation in sync with code changes and tickets. * [Customize for your organization](/agent/custom-instructions/): set standing instructions so every session follows your conventions. # Company Profile and Domains > Describe your company, website, brand names, audiences, competitors, voice and visual style once, so EkLine's agents write as your company. Company Profile is where you describe your company once: what it makes, how its names are spelled, who it writes for, who it competes with, and how it sounds. EkLine’s agents then write as your company instead of guessing. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * Any member of your organization can edit Company Profile. ## What Company Profile holds [Section titled “What Company Profile holds”](#what-company-profile-holds) The fields appear in this order under **Settings** > **Organization** > **Company Profile**: | Field | What to put there | | ---------------------------- | ------------------------------------------------------------------------------------------------------ | | **Company Profile** | What your company does, its mission and background | | **Brand names** | Each brand or product name as you write it, plus the variants to watch for, with commas between them | | **Website** | Your company’s website, such as `acme.com`. It sits with **Documentation sites** under **Your sites**. | | **Documentation sites** | The URL of each site where you publish docs, with an optional label | | **Key Features** | Your product’s main features | | **Audience (User Profiles)** | Who reads your docs and content | | **Personas** | Named readers, each with an optional note on what they are trying to do | | **Competitors** | Your main competitors, described in your own words | | **Competitor profiles** | Each competitor’s name, with an optional URL and aliases | | **Tone and Voice** | How your company sounds when it writes | | **Visual Style Guidelines** | Brand colors and the look you want for images. See [Visual style guidelines](/agent/visual-style/). | | **Additional Context** | Anything else the agents should know | The large text fields hold up to 10,000 characters each. They are **Company Profile**, **Key Features**, **Audience (User Profiles)**, **Competitors**, **Tone and Voice**, **Visual Style Guidelines** and **Additional Context**. ## What onboarding fills in [Section titled “What onboarding fills in”](#what-onboarding-fills-in) [Your first AI report](/get-started/first-ai-report/) reads your website and fills in Company Profile for you, including **Website**, **Tone and Voice** and **Visual Style Guidelines**. You check everything before EkLine saves it, and you can change any field here later. If **Audience (User Profiles)** is empty when you confirm the profile, EkLine fills it with the names of the audiences you confirmed, such as “Platform engineer, Engineering lead”. An audience you already saved is never replaced. ## Refresh Visual Style Guidelines from your website [Section titled “Refresh Visual Style Guidelines from your website”](#refresh-visual-style-guidelines-from-your-website) Under **Visual Style Guidelines**, **Refresh from website** reads the website saved under **Your sites** and drafts the colors, fonts, and logo for you. A preview shows the logo, up to four colors, and the fonts before anything changes. Choose **Replace guide** to put the draft into the field, or **Cancel** to keep your current text. If the field is empty, the button reads **Use this guide**. The field then saves like any other. The button is off until you save a website under **Your sites**. If EkLine can’t read a brand from the site, a message tells you so and nothing changes. ## Which agents read it [Section titled “Which agents read it”](#which-agents-read-it) | Agent | When it reads Company Profile | | --------------- | ---------------------------------------------------------------------------------------------------------------- | | Marketing Agent | Every session. It reads the profile before it drafts anything. | | Docs Agent | Only when company context is on for your organization. To check or change this, contact ****. | Both agents read the profile when a session starts. A change applies to the next session you start. If **Tone and Voice** is empty, the Marketing Agent says it has no company voice on file. Before it drafts, it asks you three things: who the audience is, what tone you want, and one sentence about the product. In chat, the Marketing Agent may call this page **Company Info**. It is the same page. [Journeys](/journeys/) also uses four fields. It finds your **Brand names** and **Competitor profiles** in AI answers. It uses your **Website** and **Documentation sites** to spot your own pages among the sources an answer cites. ## Steps [Section titled “Steps”](#steps) 1. Go to **Settings** > **Organization** > **Company Profile**. 2. Fill in the fields that apply to you. For the Marketing Agent, fill in **Tone and Voice** and **Brand names**. With its custom instructions, they set the tone and the names in its drafts. 3. Add a row for each brand name, documentation site, persona or competitor with **Add brand name**, **Add documentation site**, **Add persona** or **Add competitor**. 4. Each field saves on its own, so there is no button to click. A text field saves when you stop typing or move to another field, and **Saved** appears under it. A list saves once all its rows are complete, and **Saved** appears next to its add button. To remove a row, select its trash icon and confirm with **Remove**; the change saves at once. A new row you have not filled in is removed straight away. A field over 10,000 characters is marked as you type, and a row that is missing something is marked when you leave it. Neither is saved until you fix it, and a list with such a row waits until the row is fixed. If a save fails, the page shows **Couldn’t save.** with **Try again**. ## Check it worked [Section titled “Check it worked”](#check-it-worked) 1. Open the Marketing Agent and start a new session. 2. Ask for a short draft and check that it uses your product names and voice. ## Domains [Section titled “Domains”](#domains) **Settings** > **Organization** > **Domains** lists the email domains your organization claims, such as `acme.com`. When someone signs up with a verified email at a claimed domain, sign-up offers to let them join your organization. Domains is not agent context: no agent reads it. It is separate from the **Website** field. Only Owners and Admins see the tab. Anyone else who opens it sees **Only Owners and Admins can manage domain claims.** ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### The Company Profile tab isn’t there [Section titled “The Company Profile tab isn’t there”](#the-company-profile-tab-isnt-there) If you don’t see **Settings** > **Organization** > **Company Profile**, contact **** with your organization name. ### Website shows “Enter a website like acme.com” [Section titled “Website shows “Enter a website like acme.com””](#website-shows-enter-a-website-like-acmecom) Enter one public website address, such as `acme.com`. ## Next [Section titled “Next”](#next) * [Visual style guidelines](/agent/visual-style/): Set the colors and look of the images agents draw. * [Custom instructions](/agent/custom-instructions/): Add standing instructions for one agent. * [Reference documents and repository notes](/agent/reference-documents/): Upload documents and describe each repository. * [Which setting to use](/agent/which-setting-to-use/): Compare every way to give the agents context. # Create documentation with Docs Agent > Generate READMEs, API references, and guides with Docs Agent from your codebase, uploaded videos, screenshots, or external sources like Slack and Notion. Use the Docs Agent to write documentation that doesn’t exist yet. It can write from your code, from videos and images you upload, or from content in your connected tools. ## From your codebase [Section titled “From your codebase”](#from-your-codebase) The agent reads your repository’s structure, source files, package configuration, and existing docs, then writes new content from them. ### README files [Section titled “README files”](#readme-files) Generate a README when starting a new project or updating an outdated one. ```text Generate a README for this repository. Include: - Project overview and key features - Installation with npm and yarn - Quick start example - Configuration options - Contributing guidelines ``` **Tips:** * Mention specific sections you need. * Reference example code: `"Include examples similar to /examples"` * Specify the audience: `"Write for developers familiar with React"` ### API reference [Section titled “API reference”](#api-reference) Document endpoints, functions, or modules from your source code. ```text Document the authentication endpoints in src/api/auth.ts. Include request parameters, response schema, example requests, and error codes. ``` The agent extracts and documents: * Endpoint URLs and HTTP methods * Request parameters with types and descriptions * Response schema with field descriptions * Example request and response bodies * Error responses and status codes ### How-to guides [Section titled “How-to guides”](#how-to-guides) Create task-oriented guides for common workflows. ```text Create a how-to guide for setting up local development. Cover environment setup, database configuration, and running the test suite. ``` ### Getting started guides [Section titled “Getting started guides”](#getting-started-guides) Help new users or contributors get up and running. ```text Write a getting started guide for new contributors. Include repo setup, coding standards, and how to submit a pull request. ``` ## From video uploads [Section titled “From video uploads”](#from-video-uploads) Upload product demos, tutorials, or walkthroughs. The agent transcribes the video and writes docs from what it sees and hears. ### How it works [Section titled “How it works”](#how-it-works) 1. Click the attachment icon in the chat panel. 2. Select your video file. 3. Wait for processing. Transcription may take a few minutes. 4. Describe what documentation you need. ### Example prompts [Section titled “Example prompts”](#example-prompts) * Tutorial from demo ```text Create a step-by-step tutorial from this product demo. Include descriptions of each screen and action shown. ``` * Feature overview ```text Write a feature overview document based on this walkthrough video. Focus on the key capabilities demonstrated. ``` * Release notes ```text Generate release notes from this demo of new features. Summarize each feature shown for end users. ``` ## From screenshots and images [Section titled “From screenshots and images”](#from-screenshots-and-images) Upload screenshots, diagrams, or other images, and the agent adds them to your docs. ![Docs Agent drafting a documentation page from an uploaded image in the chat panel](/assets/images/docs-agent-image-upload.png) ### How it works [Section titled “How it works”](#how-it-works-1) 1. Click the attachment icon in the chat panel. 2. Select your image file. 3. Describe where and how you want the image placed. 4. The agent downloads the image into your repository and adds a markdown reference. The agent can see what the image shows, so it can write alt text and captions for it. ### Example prompts [Section titled “Example prompts”](#example-prompts-1) * Insert screenshot ```text Insert this screenshot into the getting started guide after the installation section. Add a caption describing what the user should see. ``` * Add diagram ```text Add this architecture diagram to the overview page. Place it after the "How it works" section. ``` * Multiple images ```text Add these screenshots to the deployment guide. Place each one next to the step it illustrates. ``` ## From external sources [Section titled “From external sources”](#from-external-sources) Write docs from content in your connected tools. See [Integrations](/agent/integrations) to connect them. ### From Slack threads [Section titled “From Slack threads”](#from-slack-threads) Turn support conversations and team discussions into docs. ```text Create a troubleshooting guide based on this Slack thread: https://workspace.slack.com/archives/C01234/p1234567890 ``` The agent reads the conversation, finds the problem and the fix, and writes them up as a doc. For the full Slack bot workflow, see [Turn a Slack support thread into a troubleshooting doc](/agent/slack-thread-to-troubleshooting-doc/). ### From Notion pages [Section titled “From Notion pages”](#from-notion-pages) Turn product specs, design docs, or meeting notes into docs for your users. ```text Create API documentation based on the spec in this Notion page: https://notion.so/your-workspace/api-spec-page ``` ### From Confluence pages [Section titled “From Confluence pages”](#from-confluence-pages) Pull specs or internal notes into your docs. ```text Create a deployment guide based on the runbook in this Confluence page: https://your-org.atlassian.net/wiki/... ``` ### Release notes from tickets [Section titled “Release notes from tickets”](#release-notes-from-tickets) Generate changelogs from completed Linear or Jira tickets. ```text Generate release notes for v2.1 based on Linear tickets ENG-100 through ENG-110. Organize by Features, Improvements, and Bug Fixes. Write for end users. ``` The agent reads each ticket’s title, description, and comments, and sums up each one for your users. ## Tips for better output [Section titled “Tips for better output”](#tips-for-better-output) ## Next [Section titled “Next”](#next) * [Create documentation from a demo video](/agent/create-docs-from-video/): turn a whole walkthrough into a written guide. * [Capture screenshots from a video or browsing recording](/agent/screenshots-from-video/): put a frame from a recording into a page. * [Generate an API reference from your codebase](/agent/generate-api-reference/): build reference pages from your source code. * [Update and review documentation](/agent/update-review/): change pages that already exist. # Create documentation from a demo video > Upload a product demo or walkthrough video and use Docs Agent to turn it into a review-ready guide in a pull request. Turn a product demo video into written documentation. You upload a walkthrough, and Docs Agent transcribes it and reads what happens on screen. You guide the agent to a step-by-step guide that matches your existing docs, then open it as a pull request for review. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * A repository [connected to EkLine](/agent/github-app-setup/) where the documentation lives. * A demo or walkthrough video that shows the feature you want to document. Your video must meet these limits: | Requirement | Value | | ------------ | ------------------------- | | Formats | MP4, WebM, MOV, AVI, MPEG | | Maximum size | 500 MB | ## Step 1: Prepare your demo video [Section titled “Step 1: Prepare your demo video”](#step-1-prepare-your-demo-video) A clear video produces a clear draft. Before you upload, check the following: * **Narrate the important actions.** The agent transcribes the audio, so spoken explanations become part of the source material. * **Move at a steady pace.** Pause briefly on each screen so each action is clear in the recording. * **Trim dead air.** Cut long silences or unrelated tangents to keep the transcription focused. * **Confirm the format and size.** Export as MP4, WebM, MOV, AVI, or MPEG, and keep the file under 500 MB. A three-to-ten-minute screen recording with narration works well for a single feature. ## Step 2: Open a session [Section titled “Step 2: Open a session”](#step-2-open-a-session) 1. Sign in to the [EkLine app](https://ekline.io/dashboard). 2. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. ![The Docs Agent editor with the chat panel ready for a prompt](/assets/images/docs-agent-editor.png) ## Step 3: Upload your video [Section titled “Step 3: Upload your video”](#step-3-upload-your-video) 1. Click the attachment icon in the chat panel. 2. Select your video file. 3. Wait for the upload to finish. The video appears as an attachment in the chat panel once the upload finishes. ## Step 4: Describe the documentation you want [Section titled “Step 4: Describe the documentation you want”](#step-4-describe-the-documentation-you-want) Tell the agent what to build from the video. Naming the document type and audience keeps the output focused. * Step-by-step tutorial ```text Create a step-by-step tutorial from this product demo. Describe each screen and action shown, and write it for a new user trying the feature for the first time. ``` * Feature overview ```text Write a feature overview based on this walkthrough video. Focus on the key capabilities demonstrated and who they help. ``` * Release notes ```text Generate release notes from this demo of new features. Summarize each feature shown for end users. ``` To match your existing docs, point the agent at a similar page: ```text Create a step-by-step tutorial from this demo video. Match the structure and tone of docs/tutorials/getting-started.md. ``` When you send the message, the agent first transcribes the video and reads what happens on screen, which can take a few minutes for a longer recording. It then drafts the guide in the editor panel. ## Step 5: Review the draft [Section titled “Step 5: Review the draft”](#step-5-review-the-draft) Read the draft in the editor panel and check it against the video: * The steps follow the same order as the demo. * Screen names, button labels, and menu paths match what the video shows. * The guide covers every action you demonstrated, with nothing invented. Refine anything that is off by replying in the chat. For example: ```text Add a verification step after the install section that tells the reader what a successful setup looks like. ``` The agent updates the draft in place. Repeat until the guide reads the way you want. ## Step 6: Publish as a pull request [Section titled “Step 6: Publish as a pull request”](#step-6-publish-as-a-pull-request) 1. Enable **View All Changes** to see a diff of what the agent created. 2. Click **Raise PR** to open a pull request with the new guide. 3. The chat box fills with `Open a pull request for` and your repository’s name. Press **Enter** to send it, or edit it first to add instructions. The agent opens the pull request and replies in the chat with a link where you continue the review on GitHub. ## Verify [Section titled “Verify”](#verify) Confirm the tutorial worked end to end: * [ ] The uploaded video appears as an attachment in the chat panel. * [ ] The editor panel shows a drafted guide that follows the demo. * [ ] A pull request is open on your connected repository with the new file. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | ----------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Upload is rejected | Check that the file is under 500 MB and in a supported format (MP4, WebM, MOV, AVI, MPEG). | | The agent takes a long time to respond | The agent transcribes the video after you send your message, and a longer video takes longer. Wait a few minutes, or trim the recording before you upload it. | | The draft misses steps shown in the video | Narrate the missing actions in the recording, or describe them in a chat reply so the agent adds them. | | Screen labels in the draft are wrong | Reply in the chat with the correct label and ask the agent to fix it. Attaching a screenshot of the screen helps. | ## Next steps [Section titled “Next steps”](#next-steps) * [Create documentation](/agent/create/): generate docs from your codebase, screenshots, and external sources. * [Update and review documentation](/agent/update-review/): keep the guide in sync as the feature changes. * [Generate an API reference](/agent/generate-api-reference/): turn source code into reference documentation. # Custom instructions > Write standing instructions for the Docs Agent or the Marketing Agent. Each agent reads only its own, in every new session. Custom instructions are standing guidance you write once for your organization, such as spelling, terminology, voice or audience, and an agent follows them in every new session. Each agent has its own custom instructions, and each agent sees only its own: | Agent | Where you set them | Sessions they reach | | --------------- | ------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Docs Agent | **Settings** > **Docs Agent** > **Custom instructions** | Every new Docs Agent session: started in the editor, from a [GitHub pull request](/agent/github-integration/), a [GitLab merge request](/agent/gitlab-integration/), [Slack](/agent/slack-bot/), Discord, a Jira ticket, a Journeys report, an [MCP client](/agent/mcp/), or an [automation](/agent/scheduled-agents/) | | Marketing Agent | **Settings** > **Marketing Agent** > **Custom instructions** | Every new Marketing Agent session, started in the editor, from an MCP client, from a Journeys report, or by an [automation](/agent/scheduled-agents/) that runs as the Marketing Agent | The Docs Agent never reads the Marketing Agent’s instructions, and the Marketing Agent never reads the Docs Agent’s. To give both agents the same guidance, write it in both places, or put facts about your company in [Company Profile](/agent/company-profile/), which both agents can read. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * Any member of your organization can edit custom instructions. ## Steps [Section titled “Steps”](#steps) 1. Open the settings for the agent: * Docs Agent: **Settings** > **Docs Agent** > **Custom instructions** * Marketing Agent: **Settings** > **Marketing Agent** > **Custom instructions** 2. Write your guidance in the field (**Custom Instructions** for the Docs Agent, **Marketing Agent custom instructions** for the Marketing Agent). The field holds up to 20,000 characters, and the counter below it shows how many you have used. 3. Click **Save custom instructions**. The message **Saved. It applies to sessions started from now on.** confirms the save. Your instructions apply to every session started after you save. There is nothing else to do: no refresh, no sync. A session that is already open keeps the instructions it started with. ## What to include [Section titled “What to include”](#what-to-include) Write custom instructions the way you would brief a new writer joining your team. Focus on what the agent can’t get from your code, docs or company profile. For the Docs Agent: * **Style preferences.** Spelling, capitalization, and voice. For example, `"Use American English spelling and sentence case for headings."` * **Terminology.** Terms to prefer or avoid. For example, `"Refer to the product as Acme, never ACME or acme."` * **Documentation structure.** How your docs are organized. For example, `"Place how-to guides in docs/guides and reference material in docs/reference."` * **Versions and audience.** Which versions to document and who reads the docs. For example, `"Document v2 of the API only. Write for backend developers familiar with REST."` For the Marketing Agent: * **Voice and person.** For example, `"Write in the first person plural."` * **How posts open and close.** For example, `"Open blog posts with the problem, not the product."` * **Words and claims to avoid.** For example, `"Never call a feature new unless it shipped this quarter."` The Marketing Agent takes tone, person, formality and names from its custom instructions together with the **Tone and Voice** and **Brand names** fields in [Company Profile](/agent/company-profile/). ### Examples [Section titled “Examples”](#examples) * Docs: style and terminology ```text Use American English spelling and sentence case for headings. Refer to the product as Acme. Use "sign in" as a verb and "sign-in" as an adjective. Write in the second person and active voice. ``` * Docs: structure and audience ```text Follow the Diataxis framework: tutorials, how-to guides, reference, and explanation each live in their own folder. Write for developers new to Kubernetes. Document v2 of the API only. ``` * Marketing: voice ```text Write in the first person plural. Open blog posts with the problem, not the product. Keep sentences short. Name Hooli only when a source compares us directly. ``` To describe what one repository is for, add a [repository note](/agent/reference-documents/) instead. For a comparison of every setting, see [Which setting to use](/agent/which-setting-to-use/). ## How custom instructions relate to style learning [Section titled “How custom instructions relate to style learning”](#how-custom-instructions-relate-to-style-learning) For the Docs Agent, custom instructions work alongside its automatic style learning. It still analyzes your existing documentation for heading structure, code example conventions, terminology, and file organization. Custom instructions add the context the agent can’t infer from your docs. For more on how the agent learns your style, see [How does the agent know my documentation style?](/agent/reference/#how-does-the-agent-know-my-documentation-style) in the reference. ## Check it worked [Section titled “Check it worked”](#check-it-worked) 1. Start a new session with the agent whose instructions you changed. 2. Ask for something your instructions affect. For example, ask the Docs Agent to `"Generate a README for this repository."`, or ask the Marketing Agent for a short LinkedIn post. 3. Check the draft against your instructions: spelling, heading case, terminology, voice and structure. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### A Custom instructions tab isn’t there [Section titled “A Custom instructions tab isn’t there”](#a-custom-instructions-tab-isnt-there) If you don’t see **Settings** > **Docs Agent** > **Custom instructions** or **Settings** > **Marketing Agent** > **Custom instructions**, contact **** with your organization name. ### The agent ignores an instruction [Section titled “The agent ignores an instruction”](#the-agent-ignores-an-instruction) * Check that you saved the instruction in the settings for the agent you are using. Each agent reads only its own. * Start a new session. A session that was already open keeps the instructions it started with. * Make the instruction more specific and concrete. Short, direct instructions work better than long, general ones. * For a long convention, point to a file in your repository instead of describing it inline. ## Next [Section titled “Next”](#next) * [Company Profile and Domains](/agent/company-profile/): Describe your company once for both agents. * [Reference documents and repository notes](/agent/reference-documents/): Upload documents and describe each repository. * [Skills](/agent/custom-skills/): Turn a procedure your team repeats into a command anyone can run. * [Which setting to use](/agent/which-setting-to-use/): Compare every way to give the agents context. # Package a workflow as a skill > Author a Docs Agent skill so your team can run a repeatable documentation workflow with a slash command. A skill packages a **workflow your team repeats**, such as your release-notes procedure, your API-reference checklist, or the steps you follow to turn a support thread into a troubleshooting doc, so anyone can run it in a session by typing a slash command. Each skill has three parts: a **name**, a description of **when to use it**, and the **instructions** the agent follows. You write it once for your organization, and both agents, the Docs Agent and the Marketing Agent, can use it in every new session. You run it on demand by typing its command, and the agent also reaches for it on its own when a task matches the skill’s description. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled * At least one repository [connected to EkLine](/agent/github-app-setup/) Any member of your organization can create and edit skills. ## Create a skill [Section titled “Create a skill”](#create-a-skill) 1. Log in to your [EkLine dashboard](https://ekline.io/dashboard). 2. Go to **Settings** > **Docs Agent** > **Skills**. 3. Click **Add skill**. 4. In the **Name** field, enter a short name for the workflow, such as `Release notes checklist`. As you type, the hint under the field shows the slash command you use to run the skill, for example `/release-notes-checklist`. You can enter up to 100 characters. 5. In the **When to use it** field, describe the situations the skill applies to, such as `Use when preparing release notes for a new product version.` The agent reads this to decide when to reach for the skill on its own, so be specific. You can enter up to 1,000 characters. 6. In the **Instructions** field, write the steps the agent should follow, in order. You can enter up to 50,000 characters. 7. Click **Save skill**. EkLine adds the skill to your list, where its command appears next to its name. The skill applies to every new session your organization starts after you save it. A session that is already open doesn’t pick up a skill you create or change. You can start a session in the editor or trigger the skill from a [GitHub pull request](/agent/github-integration/), [GitLab merge request](/agent/gitlab-integration/), or [Slack](/agent/slack-bot/). ### How the command is named [Section titled “How the command is named”](#how-the-command-is-named) The agent runs your skill when you type its slash command. EkLine builds the command from the skill’s name: it lowercases the name, turns accented letters into plain ones, replaces spaces and punctuation with hyphens, and keeps the first 60 characters. There is no prefix. `Release notes checklist` becomes `/release-notes-checklist`. EkLine rejects two kinds of name, because a built-in skill would take the command instead of yours: * A name whose command starts with `/ekline-`. That prefix is reserved for EkLine’s built-in skills. * A name whose command matches one of EkLine’s built-in commands. Because the command follows the name, renaming a skill changes its command. Tell your team when you rename one they use. ## Write effective instructions [Section titled “Write effective instructions”](#write-effective-instructions) Write instructions the way you’d hand a runbook to a teammate doing the task for the first time. The agent has your repositories, your [custom instructions](/agent/custom-instructions/), and its connected integrations available while it works. Your skill supplies the procedure to follow. * **Number the steps.** Order the work so the agent moves from gathering sources to drafting to review. * **Name your sources.** Point the agent at the files, integrations, or pages each step draws on. * **State the output.** Describe the document to produce, where it belongs, and the format it follows. * **Set the guardrails.** Call out what to include, what to leave out, and when to stop and ask. - Release notes ```text Draft release notes for the version the user names. 1. Read the merged pull requests since the last release tag. 2. Group the changes into Added, Changed, Fixed, and Removed. 3. Write one user-facing sentence per change in our house style. 4. Open the release notes page and add the new version at the top. 5. Leave internal refactors and dependency bumps out. ``` - API reference check ```text Check that our API reference matches the code. 1. Read the endpoint definitions in the api-gateway repository. 2. Compare each one against its entry in docs/reference/api. 3. Flag endpoints that are missing, renamed, or have changed parameters. 4. Draft the updates and list every change you made. ``` ## Run a skill [Section titled “Run a skill”](#run-a-skill) 1. Open the editor and start a new session. 2. Type the skill’s command, for example `/release-notes-checklist`, and add any detail the workflow needs, such as the version to document. 3. The agent follows your instructions and reports what it did. You don’t have to type the command every time. The agent reads each skill’s **When to use it** description and can choose a skill when your request matches. Ask for release notes in plain language, and the agent applies your release-notes skill. ## Check it worked [Section titled “Check it worked”](#check-it-worked) * The **Skills** tab lists the skill with its command next to its name, and its toggle is on. * In a new session, typing the command makes the agent follow your steps. ## Manage your skills [Section titled “Manage your skills”](#manage-your-skills) The **Skills** tab lists every skill your organization has, with its name, its command, and its **When to use it** description. To see a skill’s instructions, click its edit icon. This tab is the only place to see what a skill contains: the editor has no list of skills, and EkLine’s built-in skills are not listed anywhere in the app. For each skill, you can: * **Edit**: Click the edit icon to change the name, description, or instructions, then click **Save skill**. * **Enable or disable**: Use the toggle to enable or disable a skill. A disabled skill stays in your list but doesn’t reach new sessions, so its command doesn’t run. Disable a skill while you’re revising it, or to retire it without losing the instructions. * **Delete**: Click the delete icon to remove a skill. It disappears from your next session. ## How skills relate to custom instructions [Section titled “How skills relate to custom instructions”](#how-skills-relate-to-custom-instructions) Skills and [custom instructions](/agent/custom-instructions/) solve different problems, and most organizations use both: * **Custom instructions** shape *how* the agent writes (spelling, terminology, repository context) and apply to every session automatically. * **Skills** package *a workflow you repeat*: a named procedure you run on demand with a command, or that the agent runs when the task matches. Put a convention that should hold across all writing in custom instructions. Put a multi-step task you run again and again into a skill. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### The Skills tab isn’t available [Section titled “The Skills tab isn’t available”](#the-skills-tab-isnt-available) If you don’t see **Settings** > **Docs Agent** > **Skills**, contact **** with your organization name. ### The agent doesn’t run the skill [Section titled “The agent doesn’t run the skill”](#the-agent-doesnt-run-the-skill) * Confirm the skill is enabled. A disabled skill doesn’t reach new sessions. * Start a new session. Skills apply only to sessions started after you save. * Check the command. Type it exactly as it appears next to the skill’s name. ### The agent runs the wrong skill, or none, from a plain-language request [Section titled “The agent runs the wrong skill, or none, from a plain-language request”](#the-agent-runs-the-wrong-skill-or-none-from-a-plain-language-request) The agent chooses a skill by matching your request against each skill’s **When to use it** description. Make the description more specific about when the skill applies, or type the command directly to run the skill you want. ## Next steps [Section titled “Next steps”](#next-steps) * [Custom instructions](/agent/custom-instructions/): Add custom instructions that apply to every session. * [Automations](/agent/scheduled-agents/): Run documentation tasks on a recurring schedule. * [Docs Agent reference](/agent/reference/): Supported file types, limits, and FAQ. # Docs Agent Discord bot > Start documentation drafts from Discord by mentioning EkLine in a server thread and reading the reply in the same thread. Create documentation without leaving Discord. Mention the EkLine bot in any channel or thread on a connected server, and Docs Agent replies in the same thread with a draft you can review and edit in EkLine. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) Before you begin, you need: * An EkLine organization account. * Discord server administrator access to authorize the app. ## Connect your Discord server [Section titled “Connect your Discord server”](#connect-your-discord-server) 1. Go to **Settings > Integrations** in your EkLine dashboard. 2. Find **Discord** in the integrations list. 3. Click **Connect**. 4. On the Discord page that opens, authorize EkLine to access your Discord server. 5. Select the server to connect, then complete the authorization flow. Discord returns you to EkLine. 6. Verify the integration shows as **Connected**. Once connected, anyone in the server can mention the EkLine bot to start a draft in your EkLine organization. ## Create documentation from Discord [Section titled “Create documentation from Discord”](#create-documentation-from-discord) 1. In any channel or thread on the connected server, mention the EkLine bot followed by your request. ```text @EkLine create a troubleshooting guide based on this discussion ``` 2. Read the acknowledgment. The bot confirms it has your request, then works on the draft. It posts no progress updates, only the finished answer. 3. Read Docs Agent’s reply in the same thread. Long answers arrive as more than one message because Discord limits the length of a single message. ## Review and edit in EkLine [Section titled “Review and edit in EkLine”](#review-and-edit-in-ekline) Each Discord mention starts a Docs Agent session in the EkLine app. Open it to review the draft, refine it, and open a pull request. To find it, filter the session list by the **Discord** source. See [Find your sessions](/agent/find-sessions/). ## Context the bot captures [Section titled “Context the bot captures”](#context-the-bot-captures) | Context type | What the bot captures | | --------------- | ------------------------------------------------------- | | Thread messages | Up to the last 20 messages in the thread. | | Your request | The text of your mention, with the bot mention removed. | ## Write effective prompts [Section titled “Write effective prompts”](#write-effective-prompts) The quality of your prompt determines the quality of the output. | Instead of… | Try… | | ---------------- | ------------------------------------------------------------------------- | | “help with docs” | “Create a troubleshooting guide for the API timeout issue discussed here” | | “document this” | “Write a how-to guide for the workaround Sarah described” | | “make docs” | “Update the authentication docs based on the new flow we agreed on” | **Include specific details:** * Reference the type of documentation you need: tutorial, how-to guide, reference, troubleshooting. * Name specific topics or features from the conversation. * Mention existing documentation to update if applicable. * Point to URLs or ticket IDs for more context. ## Check it worked [Section titled “Check it worked”](#check-it-worked) * The bot posts an acknowledgment in the thread, and later Docs Agent’s reply. * The session appears under **Sessions** in the EkLine app. Filter by **Source** > **Discord** to find it. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | ------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------- | | Bot replies that it works inside servers connected to EkLine | You mentioned it in a direct message. Mention the bot in a channel or thread on a connected server. | | “This server isn’t connected to EkLine” | Ask a server administrator to connect the server in **Settings > Integrations**. | | “This server is connected, but its EkLine organization has no API token” | Contact ****. | | “I couldn’t reach EkLine just now” | Wait a moment and mention the bot again. | | Reply arrives as several messages | Discord limits message length. Read the messages in order. | | Draft does not match expectations | Give more specific instructions in your mention. Include the documentation type and key topics to cover. | ## Next steps [Section titled “Next steps”](#next-steps) * [Docs Agent Slack bot](/agent/slack-bot/): Create documentation drafts from Slack channels and threads. * [Find your sessions](/agent/find-sessions/): Open, filter, and manage every Docs Agent session, including sessions started from Discord. * [Create documentation](/agent/create/): Generate READMEs, API references, and guides from your codebase. * [Integrations](/agent/integrations/): Connect Notion, Linear, Jira, and other tools. # Crop, redact, and annotate documentation images > Use Docs Agent to make a raw screenshot publishable: crop it to the element it shows, cover personal information, and add highlights, arrows, and callouts. Ask Docs Agent to crop a documentation image to what the page is about, cover information that must not reach public documentation, or mark up the part of the interface your prose refers to. You can ask for each edit on its own or combine them in a single pass. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * A repository [connected to EkLine](/agent/github-app-setup/) that holds both your documentation pages and their image files. * At least one image the agent can reach as a file. That means an image already committed to your connected repository, one the agent captured itself from a [sandbox](/agent/sandbox/), or a frame it [extracted from a video](/agent/screenshots-from-video/). The agent edits PNG and JPEG files. It leaves vector images such as SVG untouched. ## Crop an image to what it shows [Section titled “Crop an image to what it shows”](#crop-an-image-to-what-it-shows) Cropping an image to the element it shows is the edit that helps most screenshots. 1. Open the Docs Agent editor and start a new session. 2. Name the image and what it should show: ```text Crop the screenshot in our billing guide so it shows only the Payment method panel, not the whole page. ``` 3. Select the edited image’s tab in the editor to review it. The image renders in the editor panel, and you can click it to open it full size. 4. Click **Raise PR** to open a pull request with the cropped image. The button fills the chat box with a message. Send it, and the agent opens the pull request. The agent matches the padding, width, and light or dark theme of the images already beside it on the page, so a recropped image doesn’t stand out from its neighbors. It cuts what doesn’t illustrate the point, such as unrelated sidebars, panels, and toolbars, while keeping the page header or navigation item a reader needs to orient. ## Hide personal information [Section titled “Hide personal information”](#hide-personal-information) Screenshots taken in a real account show real data: names, email addresses, phone numbers, billing details, API tokens, and faces. Hide it before the image reaches public docs. 1. Tell the agent which image and what to hide: ```text The dashboard screenshot in our getting-started guide shows a real customer's name, email, and account ID. Hide those before we publish it. ``` 2. Tell the agent whether the fields should look empty or hold sample data. Use sample data where a filled-in field is part of what the image teaches. An empty field is fine for an input that would show placeholder text: ```text Replace them with sample data rather than leaving the fields empty. ``` 3. Review each covered region in the result. Confirm nothing readable remains, including partial text at the edge of a box and the same value repeated elsewhere in the image, such as an email that also appears in a header. Covering is not blurring The agent covers information with a solid, opaque fill. It doesn’t blur or pixelate it, because blurred and pixelated text can be recovered. A blur is acceptable only on a real person’s face, where no text is at risk, and even there the agent prefers a solid placeholder. If the agent cannot fully cover a region, such as text that runs under a graphic, it still delivers the image. It flags the problem in its summary and in the pull request. Read that warning before you merge. ## Add a highlight, arrow, or numbered callout [Section titled “Add a highlight, arrow, or numbered callout”](#add-a-highlight-arrow-or-numbered-callout) Use annotations when a step refers to one control on a crowded screen. 1. Describe what to mark and how: ```text In the settings screenshot, outline the Save button and add an arrow pointing at the Advanced options toggle. ``` 2. For an interface with several parts, ask for numbered markers and refer to those numbers in your text: ```text Add numbered markers to the editor screenshot: 1 on the file tree, 2 on the chat panel, 3 on the preview pane. ``` 3. Review the placement. The agent matches the shape and accent color of annotations on the sibling images beside it, so a new image looks like it belongs. The agent puts markers on the image, not text. Explain each marker in your prose: text drawn on an image can’t be found by site search, read by screen readers, or translated. Annotations are repeatable. Asking for the same highlight twice doesn’t stack two outlines on top of each other. ## Combine edits in one pass [Section titled “Combine edits in one pass”](#combine-edits-in-one-pass) Ask for several edits together, and the agent applies them in a safe order: cover information first, annotate second, crop last. ```text Take the account settings screenshot in our onboarding guide: hide the customer's name and email, highlight the Notifications toggle, and crop it to the settings panel. ``` ## What the agent changes and what it leaves alone [Section titled “What the agent changes and what it leaves alone”](#what-the-agent-changes-and-what-it-leaves-alone) | The agent | Detail | | ------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Writes to the image’s existing path | The edited image replaces the file your page already references, so no link needs updating | | Leaves your repository untouched until you merge | The change arrives as a pull request you review, like any other documentation change | | Matches its neighbors | Padding, width, theme, annotation shape, and accent color follow the sibling images beside it | | Asks when a value might be real | In a session, if it can’t tell whether a name or number is genuine or sample data, it asks you. An unattended run covers the value instead of asking | | Leaves prose alone | It edits pixels only. For documentation that describes the interface incorrectly in text, use [Update and review](/agent/update-review/) | | Skips vector images | SVG diagrams and logos stay untouched | | Follows your call in a session | Whether an image needs cropping, covering, or annotating is yours to decide: a capture from a demo account often needs none of it. A scheduled run given no instruction defaults to making the image publishable | ## Verify the result [Section titled “Verify the result”](#verify-the-result) Before you merge the pull request: * [ ] Every covered region is fully opaque, with no readable text at its edges. * [ ] No value you asked to hide survives elsewhere in the same image. * [ ] The cropped image still shows enough context for a reader to locate the element in the live product. * [ ] Annotations point at the controls your prose names, and the numbering matches your steps. * [ ] The image sits comfortably beside the others on the page in width and theme. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | -------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The agent says it can’t find the image | Give it the path or the page that references the image. An image you attached to a chat message isn’t a file the agent can edit. Commit it to your repository first. | | Nothing happened to an SVG | Cropping and covering are raster operations. Export the diagram to PNG, or edit the SVG source directly. | | The edit covered too much or too little | Describe the region by what it contains rather than by position: “the panel with the Save button”, not “the top right”. | | Text is still faintly visible | Ask the agent to extend the covered area past every edge of the text. Report any region it says it couldn’t fully cover. | | The new image looks different from the others on the page | Name the image you want it to match: “make it the same width and theme as the other screenshots in this guide”. | | The screenshot you captured shows test data you don’t want published | Point the [sandbox](/agent/sandbox/) at an account with presentable data and recapture, rather than covering every field afterward. | ## Next steps [Section titled “Next steps”](#next-steps) * [Keep documentation screenshots up to date](/agent/refresh-doc-screenshots/): recapture stale images automatically, on demand and on a schedule. * [Browse authenticated pages with a sandbox](/agent/sandbox/): give the agent the sign-in variables it needs to capture pages behind a login. * [Capture screenshots from a video](/agent/screenshots-from-video/): turn a moment in a recording into a documentation image. * [Manage a knowledge base](/agent/manage-knowledge-base/): add and replace images on Confluence pages. # Work in the editor > Start a Docs Agent or Marketing Agent session, review what the agent drafts, and send the result on as a pull request or a copy. The editor is where you give the Docs Agent or the Marketing Agent a task, review what it drafts, and send the result on. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * For Docs Agent pull requests, a documentation repository connected through [GitHub](/agent/github-app-setup/) or [GitLab](/agent/gitlab-setup/). ## Steps [Section titled “Steps”](#steps) 1. **Start a session.** Click **New session** at the top of the sidebar. On the **Docs** or **Marketing** tab it opens that agent, and on **Home** the agent you last started a session with in this browser. If the composer shows **Docs** and **Marketing**, select the agent you want. You can also type into **Start a conversation** on the **Dashboard**. If you have both agents, select **Docs** or **Marketing** there first. 2. **Describe the work.** Type the request and send it, for example `Add a section on webhook retries to docs/webhooks.md` for the Docs Agent, or `Write a blog post about our new audit log` for the Marketing Agent. A button under the composer, such as **Draft a how-to guide** or **Write a blog post**, puts its opening words in the composer for you to finish; it does not send. To add files, click the attachment icon. The agent asks in the chat for anything else it needs. See [Marketing Agent](/marketing-agent/). 3. **Review the drafts.** See [Review changes](#review-changes). 4. **Send the result on.** See [Send the result on](#send-the-result-on). To return to an earlier session, see [Find a past session](/agent/find-sessions/). ### What the editor shows [Section titled “What the editor shows”](#what-the-editor-shows) Until the agent writes a file, the chat fills the page. Then the editor panel opens on the left, with one tab per file, and the chat moves to the right. The **Home**, **Docs**, and **Marketing** tabs above the **Sessions** list appear when your organization has both agents. Once files exist, shortcuts under the chat box, such as **Fact Checker** for the Docs Agent or **Claims check** for the Marketing Agent, add a ready-made request to the box without sending it. Select text in a draft that is not in review and it is attached to your next message, so you can ask about that passage. ### Review changes [Section titled “Review changes”](#review-changes) When a turn changes a file that already existed, the file opens in review. A **Reviewing changes** banner offers **Accept All** and **Reject All**, and each change has its own check mark (**Accept change**) and cross (**Reject change**). A file the agent created from nothing opens ready to edit. Edit drafts directly; your edits save automatically. In the Docs Agent, switch on **View All Changes** to see every change in the session against the repository. ### Send the result on [Section titled “Send the result on”](#send-the-result-on) | Agent | Control | What it does | | --------------- | -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Docs Agent | **Raise PR** | Fills the chat box with `Open a pull request`. Press Enter, or add instructions first. Grayed out with “No changes to create a PR” until the session has changes. | | Docs Agent | **View PR #N** | Replaces **Raise PR** once the pull request is open. A badge shows **Open**, **Draft**, **Closed**, or **Merged**. | | Docs Agent | **Update PR** | Appears when the session has changes the open pull request does not, and no file is in review. Fills the chat box with a request to push them. | | Docs Agent | **Update KB** | Replaces **Raise PR** for drafts in a connected knowledge base, and publishes the edits back to it. | | Docs Agent | **Linked PRs** | Replaces the single button when a session changes several repositories, with an action for each. | | Marketing Agent | **Copy** | Copies the open draft. | | Marketing Agent | **Download** | Saves Markdown, or HTML from the arrow beside it. Several drafts, or drafts with images, download as one ZIP file. | The Marketing Agent has no **Raise PR**: to put a draft in a documentation repository, ask it in the chat. For what a Docs Agent pull request contains, see [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/). ### Control the agent while it works [Section titled “Control the agent while it works”](#control-the-agent-while-it-works) | Button | When | What it does | | ------ | ------------------------------------------------- | -------------------------------------- | | Send | The agent is idle | Sends your message. | | Stop | The agent is working and the box is empty | Stops the current turn. | | Queue | The agent is working and you have typed a message | Holds the message until the turn ends. | A queued message shows **Queued · click to edit**. Click it to edit, or its cross to cancel. When the turn ends, queued messages are sent together. ### Watch the agent browse [Section titled “Watch the agent browse”](#watch-the-agent-browse) When the agent opens a browser, for example to capture a screenshot, a read-only panel opens at the bottom right. Its header reads **Live browser** while the agent browses and **Recording** when the panel replays the session afterward. You can move, resize, hide, or close it. To let the agent reach pages behind a sign-in, set up a [Sandbox](/agent/sandbox/). ## Check it worked [Section titled “Check it worked”](#check-it-worked) * Your session appears under **Sessions** in the sidebar. * In a Docs Agent session, the toolbar shows **View PR #N** after the agent opens the pull request. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | What to do | | -------------------------------------------- | -------------------------------------------------------------- | | **Update PR** does not appear after you edit | A file is still in review. Accept or reject its changes first. | ## Next [Section titled “Next”](#next) * [Ship your first docs pull request](/agent/getting-started/) * [Start work from Jira tickets](/agent/jira-triggers/) * [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/) # Find a past session > Search your Docs Agent history by keyword or paste a link to find and resume any session. Every prompt you send Docs Agent runs in a session. Search finds any session in your organization’s history, including work you never turned into a pull request, so you can pick up where you left off. You can find a session two ways: * **Search by keyword:** match on what was said in the conversation. * **Paste a link:** jump from a pull request, merge request, or ticket to the session behind it. You can also [filter the sessions list](#filter-the-sessions-list) by source or status. ## Open search [Section titled “Open search”](#open-search) 1. Open the editor. The **Sessions** section in the sidebar lists your sessions. 2. Press `⌘K` (`Ctrl+K` on Windows and Linux) to open search, or click the **Search sessions** button at the top of the sidebar. 3. Type your query. Results open in a focused view centered on the page. The **Sessions** list stays as it is while you search. Results appear in the focused view, not by filtering the list underneath. To close the view, press **Escape**, click outside it, or select a result. Your query stays in the search field so you can refine it. ## Search by keyword [Section titled “Search by keyword”](#search-by-keyword) Type at least three characters to search. Search matches your query against: * The session title * The title of the documentation pull request the session produced * The conversation: both your messages and the agent’s replies Results are ranked best match first. Each result shows the session title and a snippet of the matching text, with the matched words highlighted. An attribution label tells you where the match came from: | Label | Meaning | | ----------------- | ------------------------------------------- | | **you asked** | The match is in one of your messages. | | **agent replied** | The match is in one of the agent’s replies. | Select a result to open that session in the editor. ## Find a session from a link [Section titled “Find a session from a link”](#find-a-session-from-a-link) Paste a URL into the search field to find the session it relates to. Search resolves these links: | Paste this | Finds | | -------------------------- | ---------------------------------------------------------------------------- | | A GitHub pull request URL | The session triggered by that pull request, or the session that produced it | | A GitLab merge request URL | The session triggered by that merge request, or the session that produced it | | A Jira ticket URL | The session linked to that ticket | | An EkLine session URL | That session directly | Link search matches exactly, so a single paste takes you to the related session or sessions. Search shows the matches and lets you choose one. It never opens a session for you. You can paste a messy URL. A link that points at a pull request’s **Files** tab or a specific review comment still resolves. Self-managed hosts such as GitHub Enterprise and self-hosted GitLab work the same as github.com and gitlab.com. ## Understand your results [Section titled “Understand your results”](#understand-your-results) The header of the search view reports how many results it shows and how many sessions matched, for example `50 of 214 sessions · best match first`. The view shows up to 50 results. The second number counts everything that matched, not only the sessions already loaded in the **Sessions** list. * **No matches:** when nothing matches, the view says so and names your query. * **Older sessions not searched:** for a large history, keyword search covers the most recent sessions and shows a notice when older ones fall outside that range. Add more distinctive words to your query to surface the session you want. ## Filter the sessions list [Section titled “Filter the sessions list”](#filter-the-sessions-list) Click **Filter sessions** (the filter icon next to the **Sessions** heading) to narrow the list: * **Quick views**: **Needs attention** shows sessions that an integration or a schedule started and that are not marked done, leaving out pull request reviews where the agent found no docs update was needed. **Started by me** shows sessions you started. One quick view applies at a time. * **Source**: shows sessions by where they started: **Chat**, **GitHub**, **GitLab**, **Slack**, **Discord**, **Scheduled**, **Jira**, or **Journeys**. You can select more than one. The menu lists only the sources among the sessions already loaded. * **Hide done**: hides sessions marked as done. To mark one, open its actions menu in the list and click **Mark as done**. Filters apply to the sessions already loaded in the list, not to your whole history. To search your whole history, use [search by keyword](#search-by-keyword). Click **Clear all** to remove every filter. ## Next steps [Section titled “Next steps”](#next-steps) * [Getting started with Docs Agent](/agent/getting-started): Create your first document in the editor. * [Update and review](/agent/update-review): Resume a session to keep docs in sync with code changes. # For engineers: an EkLine pull request or comment on your repository > Why EkLine commented on your pull request or opened a docs pull request, how to reply, and how to tune or silence it. If EkLine commented on your pull request or asked you to review a docs pull request, you can reply to it, ask it for changes, or have its settings changed. ## Why it is there [Section titled “Why it is there”](#why-it-is-there) Someone in your organization connected this repository to EkLine, so the Docs Agent can check code changes for their effect on your documentation. What you see depends on how it was started: | What you see | Why | | ------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | One comment from EkLine on your pull request | Automatic PR review looked at your change. By default the agent posts nothing when no docs update is needed, a two or three line note when an update might be needed, and, when one is clearly needed, a comment that starts with `@ekline-ai` and starts the Docs Agent on the update. It posts at most one comment per pull request. | | “EkLine Docs Agent is on it. Follow the session in EkLine Editor.” (on GitLab: “EkLine Docs Agent is working on this update.”) | Someone mentioned `@ekline-ai`, or EkLine’s own comment did. The Docs Agent is working on the request. | | A pull request titled `docs: ...`, with you as a reviewer | The Docs Agent drafted the docs change in your documentation repository. You were added because you wrote the source pull request, or because you are on the team’s default reviewer list. | Automatic PR review runs when a pull request is opened or marked ready for review, and on GitHub when a label is added or removed. It does not run again when you push new commits, and it looks at each pull request at most once. By default it skips draft pull requests, pull requests into branches other than the default branch, and common bots such as Dependabot and Renovate. ## Reply or ask for changes [Section titled “Reply or ask for changes”](#reply-or-ask-for-changes) * Mention `@ekline-ai` in a comment on the pull request, or in an inline review comment on GitHub. A mention inside code, a link or an email address does not count. * Ask a question and the Docs Agent answers in the thread. Ask for a docs change and it makes the edits and, by default, opens or updates a docs pull request. * To change EkLine’s own docs pull request, mention `@ekline-ai` in a comment on it. On GitHub this continues the session that opened it. On GitLab it starts a new session. * Review, merge or close the docs pull request as you would any other. To get a reply, you need write or admin access to the repository, or membership of the GitHub organization. On GitLab you need the Developer role or higher. See [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/) for what a docs pull request contains. ## Tune it or turn it off [Section titled “Tune it or turn it off”](#tune-it-or-turn-it-off) These settings are in EkLine, not in your repository. Ask the person who set EkLine up, or any member of your EkLine organization, to change them. | You want to | Change this | | ------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Skip some pull requests, such as release branches, a bot, a `no-docs` label, or changes that only touch test files | The repository’s [review rules](/agent/review-rules/): **Settings** > **Integrations** > **GitHub Integration** (or **GitLab Integration**), then the **Review rules** icon on the repository. | | Stop reviews of this repository | Turn off **Enable monitoring** on the repository, or remove it from both repository lists. | | Keep the comments but stop automatic docs pull requests | Turn off **Auto-create documentation PRs** in the repository’s review rules. If you don’t see it, contact . | | Stop being added as a reviewer | **Settings** > **Docs Agent** > **Reviewers**: remove the name from **Default reviewers**, or turn off **Auto-assign reviewer**. | | Stop scheduled docs pull requests | In **Automations**, turn the automation’s card off, or clear **Raise a pull request** in its settings. See [Automations](/agent/scheduled-agents/). | | Stop automatic PR review for the whole organization | Contact . | A review rule only affects reviews that have not run yet. To skip a pull request by label, add the label before the pull request is opened or marked ready for review. For more, see [Automatic PR review](/agent/automatic-pr-review/) and [GitHub pull requests](/agent/github-integration/). # Generate an API reference from your codebase > Use Docs Agent to turn your source code into a complete, accurate API reference with parameters, response schemas, and examples. Docs Agent reads your function signatures, types, and comments, and writes an API reference from them, with parameters, response schemas, example requests, and error codes. You review the reference and open it as a pull request. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * A repository [connected to EkLine](/agent/github-app-setup/) that contains the API source code you want to document. * The path to the file or directory that defines your API, for example `src/api/auth.ts` or `src/routes/`. ## Step 1: Open a session [Section titled “Step 1: Open a session”](#step-1-open-a-session) 1. Sign in to the [EkLine app](https://ekline.io/dashboard). 2. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. ## Step 2: Point the agent at your API source [Section titled “Step 2: Point the agent at your API source”](#step-2-point-the-agent-at-your-api-source) Reference the exact file or directory that defines your endpoints. Naming the source keeps the agent focused and improves accuracy. * Single file ```text Document the endpoints in src/api/auth.ts. Include request parameters, response schema, example requests, and error codes. ``` * Directory ```text Generate an API reference for every route in src/routes/. Group the endpoints by resource, and for each one include the HTTP method, parameters, response schema, and error responses. ``` * Specific endpoints ```text Document the POST /users and GET /users/:id endpoints in src/api/users.ts. Include request and response examples with realistic values. ``` For each endpoint, the agent extracts: * Endpoint URLs and HTTP methods * Request parameters with types and descriptions * Response schema with field descriptions * Example request and response bodies * Error responses and status codes ## Step 3: Review the generated reference [Section titled “Step 3: Review the generated reference”](#step-3-review-the-generated-reference) The agent writes the reference to the editor panel as it works. Read through the output and confirm it matches your code: * Endpoint paths and methods are correct. * Every parameter your code accepts appears in the reference. * Response fields match what your code returns. ## Step 4: Refine with follow-up prompts [Section titled “Step 4: Refine with follow-up prompts”](#step-4-refine-with-follow-up-prompts) Add detail or fix gaps by continuing the conversation. The agent keeps the context from your first prompt, so you can build on the reference incrementally. * Add examples ```text Add a curl example and a JavaScript fetch example to each endpoint. ``` * Document errors ```text Add an error table to each endpoint listing the status code, error message, and cause. ``` * Match your style ```text Match the structure and tone of docs/api/payments.md. ``` ## Step 5: Raise a pull request [Section titled “Step 5: Raise a pull request”](#step-5-raise-a-pull-request) 1. Enable **View All Changes** in the toolbar to see a diff of the generated reference. 2. Make any final edits directly in the editor panel. 3. Click **Raise PR**. The chat box fills with `Open a pull request for` and your repository’s name. Press **Enter** to send it, or edit it first to add instructions. The agent opens the pull request and replies in the chat with a link to continue the review on GitHub. ## Verify the reference is accurate [Section titled “Verify the reference is accurate”](#verify-the-reference-is-accurate) Before you merge, confirm the reference reflects your code: * **Signatures match.** Every parameter name, type, and required or optional flag matches the source. * **Examples run.** Copy an example request and run it against your API. The response matches the documented schema. * **Errors are complete.** Each documented status code corresponds to an error your code actually returns. * **Nothing is missing.** Every endpoint in the referenced file or directory appears in the output. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause | Fix | | ----------------------------------------- | ------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------- | | The agent misses endpoints | The prompt was too broad, or endpoints span several files | Name each file explicitly, or reference the parent directory: `"Document every route in src/routes/"` | | Parameter descriptions are vague | Your source code lacks comments or type annotations for those parameters | Add doc comments or types in the source, then regenerate, or supply the descriptions in a follow-up prompt | | Example values look generic | The agent had no sample data to draw from | Ask for realistic values: `"Use realistic example values based on the types in the code"` | | The reference mixes in conceptual content | The prompt asked for explanation alongside reference | Keep the prompt reference-focused, and generate concepts separately with [Create documentation](/agent/create/) | ## Next steps [Section titled “Next steps”](#next-steps) * [Update and review](/agent/update-review/): keep the reference in sync when your API changes. * [Create documentation](/agent/create/): generate READMEs, guides, and more from the same codebase. * [Set up automated style checks](/reviewer/quickstart/): catch style and terminology issues on every pull request with Docs Reviewer. # Generate documentation from GitHub pull requests > Trigger Docs Agent from GitHub pull requests by mentioning @ekline-ai in PR comments or code review threads to generate documentation. Start documentation sessions without leaving GitHub. Mention `@ekline-ai` in a pull request comment or review thread, and the bot generates a documentation draft linked to the PR context. ## How it works [Section titled “How it works”](#how-it-works) When you mention `@ekline-ai` in a PR comment, the bot: 1. Reacts with a :eyes: emoji to acknowledge your request. 2. Posts a reply with a link to the EkLine editor where it creates the draft. 3. Generates documentation based on your prompt and the PR context. 4. Replies with its answer when it finishes. If it opened or updated a docs pull request, the reply includes a link to it. The bot works in pull request comments and in review comments on code lines, not in comments on issues. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) Before you begin, you need: * An EkLine organization account. * The [EkLine GitHub App](/agent/github-app-setup/) installed on your organization. * Write or admin access to the repository, or membership of its GitHub organization. The bot answers anyone else with a note that it responds only to members and collaborators. ## Trigger the bot from a PR comment [Section titled “Trigger the bot from a PR comment”](#trigger-the-bot-from-a-pr-comment) 1. Open a pull request on a repository where the EkLine GitHub App is installed. 2. Write a comment on the PR that mentions `@ekline-ai` followed by your documentation request. ```text @ekline-ai Create a migration guide for the database schema changes in this PR. ``` 3. The bot reacts with :eyes: to confirm it received your request. 4. A reply appears with a link to the EkLine editor session where the agent drafts the documentation. 5. Review and edit the draft in the EkLine editor. 6. When the agent finishes, it replies on your PR. For a documentation change, the reply links to a docs pull request in a repository under **Where EkLine writes**. If the reply has no link, open the session in the EkLine editor and click **Raise PR**. ## Trigger the bot from a code review comment [Section titled “Trigger the bot from a code review comment”](#trigger-the-bot-from-a-code-review-comment) For more targeted documentation, mention the bot on specific lines in a code review. 1. Start a review on the pull request, or view the **Files changed** tab. 2. Select the line or lines you want to reference, and add a review comment that mentions `@ekline-ai`. ```text @ekline-ai Document this new API endpoint, including the request and response schema. ``` 3. The bot reacts with :eyes: and replies in the review thread. 4. The agent reads the file path and the diff for the lines you commented on, along with your prompt. 5. Review the generated draft in the EkLine editor and the resulting docs PR. ## What the bot posts [Section titled “What the bot posts”](#what-the-bot-posts) | Stage | What the bot posts | | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Request received | :eyes: reaction on your comment | | Session started | Reply with a link to the EkLine editor | | Agent finished | A new reply with the agent’s answer, in the same review thread when you commented on code lines. If the agent opened or updated a documentation pull request, the reply links to it | | Error | Reply that reads “Sorry, I encountered an error while processing your request. Please try again later.” | If you mention the bot again on the same PR, it resumes the existing session rather than starting a new one. ## Write effective prompts [Section titled “Write effective prompts”](#write-effective-prompts) The quality of your prompt determines the quality of the output. Everything in your comment after `@ekline-ai` is the prompt. | Instead of… | Try… | | ---------------- | ---------------------------------------------------------------------------- | | “document this” | “Create an API reference for the new `/users` endpoint added in this PR” | | “update docs” | “Update the authentication guide to cover the OAuth flow changes in this PR” | | “help with docs” | “Write a migration guide for the breaking changes to the config schema” | **Include specific details:** * Name the type of documentation you need: API reference, migration guide, tutorial, how-to. * Reference specific files or changes in the PR. * Mention existing documentation to update if applicable. * Give any audience context, such as whether it’s for developer docs or an end-user guide. ## Follow-up comments [Section titled “Follow-up comments”](#follow-up-comments) After the bot creates a docs PR, mention `@ekline-ai` again to continue the conversation. You can comment on either the source pull request or the generated docs PR. The bot resumes the existing session and keeps the context of the documentation it already produced. On the docs PR, it reacts with :eyes: but posts no editor link. The agent reads each follow-up comment and decides whether you are asking a question or requesting a change: * **Questions and clarifications:** When you ask something like “Where did you get this URL?” or “Why did you change this section?”, the agent replies with an answer. It does not edit files or push commits. * **Change requests:** When you ask for a documentation change, the agent edits the docs and pushes new commits to the linked docs PR, or opens one if none exists yet. To change the documentation, describe the change you want. ## Check it worked [Section titled “Check it worked”](#check-it-worked) * A :eyes: reaction appears on your comment, followed by a reply that links to the EkLine editor. * The session appears under **Sessions** in the EkLine app. Filter by **Source** > **GitHub** to find it. See [Find a past session](/agent/find-sessions/). ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | -------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Bot does not respond | Verify that the EkLine GitHub App is installed on the repository, and that you mentioned the bot on a pull request, not an issue. | | Bot replies that it responds only to members and collaborators | Ask a repository admin to give you write access, or to add you to the GitHub organization. | | Bot reacts but does not reply | The agent may still be processing your request, and a session can take several minutes. If no reply appears, try again or contact . | | “Sorry, I encountered an error” reply from the bot | Try again. If the error repeats, contact . | | Docs PR not created | The agent opens a docs PR only for a documentation change, not for a question. If your prompt is unclear, the agent may not change any files, so give a more specific prompt. When **Auto-create documentation PRs** is off in the repository’s [review rules](/agent/review-rules/), the drafts stay in the editor. To open the PR yourself, open the session and click **Raise PR**. | ## Next steps [Section titled “Next steps”](#next-steps) * [Create documentation](/agent/create/): Generate documentation from your codebase, videos, and external sources. * [Slack bot](/agent/slack-bot/): Create documentation drafts directly from Slack. * [Integrations](/agent/integrations/): Connect Notion, Linear, Jira, and other tools for richer context. # Generate documentation from GitLab merge requests > Trigger Docs Agent from GitLab merge requests by mentioning @ekline-ai in a merge request comment to generate documentation from the change. Start docs sessions without leaving GitLab. Mention `@ekline-ai` in a merge request comment, followed by your request. Docs Agent drafts the docs from the merge request. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) Before you begin, you need: * An EkLine organization account. * GitLab connected to EkLine, with a token, your code and documentation projects selected, and the webhook configured. See [Connect GitLab](/agent/gitlab-setup/). * The webhook on your GitLab project sending **Comments** and **Merge request events** to EkLine. * The Developer role or higher on the GitLab project. Anyone else gets a note that Docs Agent responds only to members and collaborators. ## How it works [Section titled “How it works”](#how-it-works) When you mention `@ekline-ai` in a merge request comment, Docs Agent: 1. Adds a :eyes: reaction to your comment. 2. Replies with a link to the EkLine editor where it drafts the docs. 3. Writes the docs from your prompt and the merge request. 4. Edits that reply to show its answer when it finishes. For a docs change, it opens a merge request in your docs project. Docs Agent replies in the thread of the comment that mentioned it. ## Trigger Docs Agent from a merge request comment [Section titled “Trigger Docs Agent from a merge request comment”](#trigger-docs-agent-from-a-merge-request-comment) 1. **Open a merge request** on a GitLab project you connected to EkLine. 2. **Add a comment that mentions `@ekline-ai`** followed by your docs request. ```text @ekline-ai Create a migration guide for the database schema changes in this merge request. ``` 3. **Wait for the :eyes: reaction.** It confirms Docs Agent has your request. 4. **Open the editor from the reply.** The reply links to the EkLine editor session. Review and edit the draft there. 5. **Review the docs merge request.** When Docs Agent finishes, it edits its reply to show its answer. Find the merge request in your docs project, or in the editor session. If there is none, click **Raise PR** in the editor session. ## Write effective prompts [Section titled “Write effective prompts”](#write-effective-prompts) The quality of your prompt determines the quality of the output. Everything in your comment after `@ekline-ai` is the prompt. | Instead of… | Try… | | ---------------- | --------------------------------------------------------------------------------------- | | “document this” | “Create an API reference for the new `/users` endpoint added in this merge request” | | “update docs” | “Update the authentication guide to cover the OAuth flow changes in this merge request” | | “help with docs” | “Write a migration guide for the breaking changes to the config schema” | Include specific details: * Name the type of documentation you need: API reference, migration guide, tutorial, how-to. * Reference specific files or changes in the merge request. * Mention existing documentation to update if applicable. * Give any audience context, such as whether it’s for developer docs or an end-user guide. ## Follow-up comments [Section titled “Follow-up comments”](#follow-up-comments) To continue the conversation, mention `@ekline-ai` again on the source merge request. Docs Agent resumes the same session and keeps what it already wrote. A mention on the docs merge request that EkLine opened starts a new session. Docs Agent reads each follow-up comment and decides whether you are asking a question or requesting a change: * **Questions and clarifications:** When you ask something like “Where did you get this URL?” or “Why did you change this section?”, Docs Agent replies with an answer. It does not edit files or push commits. * **Change requests:** When you ask for a docs change, Docs Agent edits the docs. It pushes new commits to the linked docs merge request, or opens one if none exists yet. ## Reference GitLab content in your prompt [Section titled “Reference GitLab content in your prompt”](#reference-gitlab-content-in-your-prompt) Docs Agent reads GitLab merge request and issue links in your prompt from your connected projects. Link another merge request or an issue to give it more context: ```text @ekline-ai Document this endpoint and cross-check the request format against issue #215. ``` For the full list of what Docs Agent can reference, see [Docs Agent integrations](/agent/integrations/). ## Check it worked [Section titled “Check it worked”](#check-it-worked) * A :eyes: reaction appears on your comment, followed by a reply that links to the EkLine editor. * The session appears under **Sessions** in the EkLine app. Filter by **Source** > **GitLab** to find it. See [Find a past session](/agent/find-sessions/). ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | --------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Docs Agent does not respond | Confirm GitLab is connected and the project is selected in your integration settings. In GitLab, open **Settings > Webhooks**. Check that the webhook sends **Comments** and **Merge request events**, and that its status is healthy. | | Docs Agent replies that it responds only to members and collaborators | Ask a project maintainer to give you the Developer role or higher. | | Reaction appears but no reply | Docs Agent may still be working. A session can take several minutes. If no reply appears, mention `@ekline-ai` again or contact . | | Reply says “I couldn’t access GitLab with the stored access token” | Add a new token in [Connect GitLab](/agent/gitlab-setup/). Confirm the status shows **Token is set**, then try again. | | “Sorry, I encountered an error” reply | Try again. If the error repeats, contact . | | No docs merge request created | Docs Agent opens a merge request only for a docs change, not for a question. If your prompt is unclear, it may not change any files, so give a more specific prompt. To open the merge request yourself, open the session and click **Raise PR**. | ## Next steps [Section titled “Next steps”](#next-steps) * [Connect GitLab](/agent/gitlab-setup/): Add a token, select projects, and configure the webhook. * [GitHub PR bot](/agent/github-integration/): Trigger Docs Agent from GitHub pull requests. * [Create documentation](/agent/create/): Generate documentation from your codebase, videos, and external sources. * [Integrations](/agent/integrations/): Connect Slack, Notion, Jira, and other tools for richer context. *** ## Stuck? [Section titled “Stuck?”](#stuck) Reply to your welcome email or contact . We read every message. # How EkLine opens pull requests > Which repositories EkLine writes to and which it only reads, what a docs pull request contains, and how to ask for changes. EkLine opens pull requests (merge requests on GitLab) only in the repositories you select for documentation. Your team reviews and merges them. ## Where EkLine writes and where it only reads [Section titled “Where EkLine writes and where it only reads”](#where-ekline-writes-and-where-it-only-reads) You select repositories in **Settings** > **Integrations** > **GitHub Integration** (or **GitLab Integration**), in two lists: | List | Shown as | What EkLine does there | | ----------------------------- | ---------------- | -------------------------------------------------------------------------------------------- | | **Where EkLine writes** | **Write access** | Opens pull requests with documentation changes. Select the repositories that hold your docs. | | **What else EkLine can read** | **Read-only** | Reads code and pull requests to work out what changed and why. Never writes. | In every agent session, EkLine blocks: * edits to files outside the **Where EkLine writes** repositories * `git push` to any other repository * GitHub and GitLab command-line writes, such as creating a pull request, aimed at any other repository EkLine can still comment on pull requests in your read-only repositories. That is how [automatic PR review](/agent/automatic-pr-review/) and replies to `@ekline-ai` work. The Marketing Agent works under the same list. Its drafts stay in the editor until you ask it to publish one to a documentation repository. For EkLine’s security compliance and subprocessors, see the [trust center](https://ekline.io/trust-center). ## When EkLine opens a pull request [Section titled “When EkLine opens a pull request”](#when-ekline-opens-a-pull-request) The agent opens a pull request as a step in a session. It opens one in these cases: | Where the work started | When the agent opens a pull request | | ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | [The editor](/agent/editor/) | You click **Raise PR**, which puts “Open a pull request” in the message box, and send it. | | A comment that mentions `@ekline-ai` on a pull request or merge request | When the comment asks for a documentation change. A question gets an answer and no pull request. | | [Slack](/agent/slack-bot/) | When the request changes docs files and **Automatically raise PRs from Slack requests** is on (the default) in **Settings** > **Docs Agent** > **Slack triggers**. | | An [automation](/agent/scheduled-agents/) | When **Raise a pull request** is checked under **Notifications**. | | A [Jira trigger](/agent/jira-triggers/) | Once for each ticket the trigger picks up. | | [Automatic PR review](/agent/automatic-pr-review/) | When it rates a code pull request High, it starts the Docs Agent on the update. | | The Marketing Agent | Only when you ask it to publish a draft to a documentation repository and say where. | When **Auto-create documentation PRs** is off in a repository’s [review rules](/agent/review-rules/), comments on that repository’s pull requests leave drafts in the editor. You raise the pull request yourself. If you don’t see that setting, EkLine always opens the pull request. To get the setting, contact . If no pull request appears when you expect one, open the session and ask the agent to open one. ## What a docs pull request contains [Section titled “What a docs pull request contains”](#what-a-docs-pull-request-contains) On GitHub, EkLine opens the pull request as the EkLine GitHub App. On GitLab, it opens the merge request as the user who owns the access token you connected. A docs pull request from EkLine has: * A branch with a `docs/` prefix, based on the repository’s default branch. * A title and commit messages that start with `docs:`, followed by what was documented. * A body that says what was documented, why, and how a reviewer can check it. It also lists any of your Review rules the change could not meet. * A body that ends with ”📝 Created with EkLine · View session”, where **View session** links to the session in EkLine. * No labels. It opens ready for review. * Reviewers from **Settings** > **Docs Agent** > **Reviewers**: everyone under **Default reviewers**. When **Auto-assign reviewer** is on, it also adds the person who started the request: the Slack user who mentioned EkLine, or the author of the source pull request. If EkLine finds no reviewer, it opens the pull request anyway. Before it commits, the Docs Agent checks its changes against your rules of type **Review** (**Settings** > **Docs Agent** > **Rules**). It also runs a style review. A failed style review never stops the pull request. A pull request from the Marketing Agent has no style review. ## Confidence ratings belong to code pull requests [Section titled “Confidence ratings belong to code pull requests”](#confidence-ratings-belong-to-code-pull-requests) Low, Medium and High are the ratings [automatic PR review](/agent/automatic-pr-review/) gives a code pull request. They say how likely the change is to need a docs update. You see them in the session, as “Documentation impact confidence”, and as a dot in the sessions list. They never appear in a docs pull request. By default the agent posts nothing for Low, a short comment for Medium, and a comment that starts the Docs Agent for High. ## GitLab differences [Section titled “GitLab differences”](#gitlab-differences) * EkLine opens merge requests on gitlab.com projects. * Merge requests and replies are posted as the user who owns your organization’s GitLab access token, not as a bot. * If the access token is missing, or GitLab rejects it, the agent stops before it changes anything. It asks you to reconnect GitLab in EkLine settings. * The agent tries to add the same reviewers. A name that is not a GitLab user is left off, and the merge request opens anyway. * A comment that mentions `@ekline-ai` on EkLine’s own merge request starts a new session. It does not continue the one that opened the merge request. ## Ask for changes [Section titled “Ask for changes”](#ask-for-changes) * **In the editor:** open the session from the pull request’s **View session** link, ask for the change, and review it. Then click **Update PR**. It puts “Update pull request #N with my latest changes” in the message box. Send it. * **On GitHub:** mention `@ekline-ai` in a comment on EkLine’s pull request. EkLine reacts with 👀 and continues the session that opened the pull request, on the same branch. * **On GitLab:** mention `@ekline-ai` in a comment on the merge request. This starts a new session. In the editor and on GitHub, the agent pushes a new commit to the same pull request. It leaves the title, body and reviewers as they are. ## FAQ [Section titled “FAQ”](#faq) ### Will EkLine open pull requests on our code repositories? [Section titled “Will EkLine open pull requests on our code repositories?”](#will-ekline-open-pull-requests-on-our-code-repositories) No, unless you add a code repository to **Where EkLine writes**. Repositories under **What else EkLine can read** are read-only: EkLine blocks pushes and pull request commands aimed at them. It does comment on their pull requests, when automatic PR review covers the repository or when someone mentions `@ekline-ai`. ### Who can ask EkLine for changes in a comment? [Section titled “Who can ask EkLine for changes in a comment?”](#who-can-ask-ekline-for-changes-in-a-comment) On GitHub, someone with write or admin access to the repository, or a member of the GitHub organization. On GitLab, a project member with the Developer role or higher. Anyone else gets a reply saying EkLine responds only to members and collaborators. ### Why did EkLine not answer my second comment straight away? [Section titled “Why did EkLine not answer my second comment straight away?”](#why-did-ekline-not-answer-my-second-comment-straight-away) A session works on one request at a time. On GitHub, a later comment waits for the current one to finish, for up to five minutes. If it is still waiting after that, EkLine replies “Sorry, I encountered an error while processing your request. Please try again later.” Comment again when the first request is done. # Start work from Jira tickets > Set up a Jira trigger so a ticket that reaches a status you choose starts the Docs Agent, which drafts the docs change and opens a pull request. A Jira trigger starts a Docs Agent session when a ticket in the projects you choose, assigned to the people you choose, moves into a status you choose. The agent reads the ticket, drafts the docs change, opens a pull request, and comments on the ticket with what it did. Triggers start the Docs Agent only, never the Marketing Agent. Linear works differently: once connected, the agent can read Linear issues for context, but a Linear issue cannot start a session. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * A documentation repository connected through [GitHub](/agent/github-app-setup/) or [GitLab](/agent/gitlab-setup/). The agent opens its pull requests there. * A **Jira triggers** tab under **Settings** > **Docs Agent**. If you do not see it, contact . Any member of your organization can add, edit, and delete triggers. ## Steps [Section titled “Steps”](#steps) 1. **Connect Jira.** Go to **Settings** > **Integrations**, find **Jira**, and click **Connect**. In **Jira Subdomain**, enter the first part of your Atlassian URL (for `acme.atlassian.net`, enter `acme`), click **Connect**, and finish the Atlassian sign-in. The card then shows **Connected**. 2. **Open the triggers.** Go to **Settings** > **Docs Agent** > **Jira triggers** and click **Add trigger**. 3. **Select the tickets.** Fill in the form: | Field | What to enter | | --------------------- | ------------------------------------------------------------------------------------------------------------------- | | **Projects** | One or more Jira projects. | | **Assignees** | One or more people. Type in **Search people by name** and select from the results. | | **Statuses** | One or more statuses, for example `Ready for docs`. | | **Labels (optional)** | Leave empty, or limit the trigger to tickets that carry at least one of these labels. Type a label and press Enter. | 4. **Save it.** Click **Create trigger**. The trigger appears as a sentence, such as “When a ticket in Acme Help (ACME), assigned to Alex Kim, enters Ready for docs”, with a switch to turn it off and buttons to edit or delete it. ### How tickets match [Section titled “How tickets match”](#how-tickets-match) * **Any value within a field, every field together.** A ticket matches when its project, assignee, and status are each one of the trigger’s values, and, with labels set, it carries at least one of them. Unassigned tickets never match. * **On the way in.** The trigger fires when a change of status, assignee, or (with labels set) labels makes a ticket match, or when a ticket is created already matching. Editing the summary or description does not fire it. * **Statuses match by name**, ignoring case. If several workflows have a status called `Ready for docs`, the list shows it once and all of them count. * **One session per ticket.** Once a ticket has started a session, moving it out of the status and back does not start another. If that session failed before it finished, the next matching change starts a new one. ### What the agent does [Section titled “What the agent does”](#what-the-agent-does) 1. Checks the ticket again immediately before it starts. If the ticket no longer matches, it stops without doing anything. 2. Reads the ticket, its comments, and any linked pull requests and Confluence pages. 3. Decides which documentation repository and pages the change belongs in, from the ticket, the linked code, and your connected repositories. There is no setting for a target repository. 4. Drafts the update and opens a pull request for your team to review. 5. Posts one comment on the ticket: what it changed and why, the pull request link, and anything it could not read. If the agent cannot tell where the docs belong, or the change needs no docs, it opens no pull request and says so in its ticket comment. For what the pull request contains, see [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/). ## Check it worked [Section titled “Check it worked”](#check-it-worked) Move a test ticket into the trigger’s status, assigned to one of its assignees, then look for: * **A comment on the ticket** with the pull request link. * **A session** under **Sessions**, titled with the ticket’s summary. Filter by **Source** > **Jira**, or paste the ticket’s URL into session search. See [Find a past session](/agent/find-sessions/). * **A pull request** in your documentation repository. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | What to do | | ------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | | The tab says “Connect Jira first to configure triggers” | Connect Jira in **Settings** > **Integrations**, then reopen the tab. | | **Create trigger** stays grayed out | Select at least one project, one assignee, and one status. | | A ticket reached the status and nothing happened | Check that the trigger is switched on, the ticket’s assignee and labels match, and the ticket has not started a session before. | | The agent commented but opened no pull request | It could not tell where the docs live, or found nothing to change. Open the session and tell the agent where the docs live. | ## Next [Section titled “Next”](#next) * [Work in the editor](/agent/editor/) * [How EkLine opens pull requests](/agent/how-ekline-opens-pull-requests/) * [Slack](/agent/slack-bot/) * [GitHub pull requests](/agent/github-integration/) # Keep documentation in sync with code > Docs Agent keeps documentation in sync with code: it checks pull requests for docs impact, audits for drift on a schedule, and drafts updates for you to review. EkLine Docs Agent keeps your documentation in sync with your code. Docs drift when a pull request merges without a matching docs update, such as a renamed option, a new API endpoint, or a removed feature. The agent watches your repositories for changes, finds where the docs have fallen behind, and drafts the update as a pull request you review. ## How EkLine keeps documentation in sync [Section titled “How EkLine keeps documentation in sync”](#how-ekline-keeps-documentation-in-sync) EkLine catches drift three ways. The first two run on their own once you set them up. The third runs when you ask. | Mechanism | When it runs | What it does | | --------------------------------- | ------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Automatic pull-request review** | When a pull request opens | Checks the diff for docs impact. On a high-confidence match, it drafts the docs update and opens a docs pull request. See [Automatic PR review](/agent/automatic-pr-review/). | | **Scheduled drift audit** | On a cadence you set | Reviews recent code changes, finds customer-facing changes with no docs, and fixes the most important gap. See [Prevent documentation drift](/agent/prevent-documentation-drift/). | | **On-demand updates** | When you ask | Updates the affected docs from a merged change or a referenced Jira or Linear ticket. See [Update and review](/agent/update-review/). | ## What EkLine keeps current [Section titled “What EkLine keeps current”](#what-ekline-keeps-current) * **README files**, from your repository structure and code. * **API references**, from function signatures, types, and comments. * **How-to and getting-started guides**, for your users and contributors. * **Knowledge-base pages**, such as Confluence spaces connected to EkLine. ## Where it fits in your workflow [Section titled “Where it fits in your workflow”](#where-it-fits-in-your-workflow) EkLine connects to [GitHub](/agent/github-app-setup/) and [GitLab](/agent/gitlab-setup/) to watch code changes and open documentation pull requests. Every update arrives as a pull request on your documentation repository, so it goes through your usual review and merge. ## FAQ [Section titled “FAQ”](#faq) ### Can EkLine automatically update documentation when code changes? [Section titled “Can EkLine automatically update documentation when code changes?”](#can-ekline-automatically-update-documentation-when-code-changes) Yes. When a pull request opens, EkLine Docs Agent checks the diff for docs impact. On a high-confidence match, it drafts the docs update and opens a docs pull request for you to review. You can also point the agent at a merged change or a Jira or Linear ticket, and it updates the affected docs. ### How does EkLine detect documentation drift? [Section titled “How does EkLine detect documentation drift?”](#how-does-ekline-detect-documentation-drift) EkLine finds drift two ways. Automatic pull-request review checks each new pull request against your existing docs. It runs in the repositories you select, and skips draft pull requests by default. A scheduled drift audit reviews recent code changes on a schedule you set and fixes the most important undocumented change. It catches anything that merged before you connected EkLine or fell below the automatic threshold. ### Which platforms does EkLine work with to keep developer docs current? [Section titled “Which platforms does EkLine work with to keep developer docs current?”](#which-platforms-does-ekline-work-with-to-keep-developer-docs-current) EkLine connects to GitHub and GitLab to watch code changes and open documentation pull requests. It reads context from Jira and Linear tickets, and pulls supporting content from Slack, Notion, Confluence, and Google Drive. ### Does EkLine change my documentation without review? [Section titled “Does EkLine change my documentation without review?”](#does-ekline-change-my-documentation-without-review) No. EkLine drafts the update and opens a pull request. You review and merge it through your normal process before any change reaches your main branch. ### What kinds of documentation can EkLine keep in sync? [Section titled “What kinds of documentation can EkLine keep in sync?”](#what-kinds-of-documentation-can-ekline-keep-in-sync) README files, API references, how-to guides, getting-started guides, and knowledge-base pages. EkLine reads your existing docs and matches their style, terms, and structure. ## Next steps [Section titled “Next steps”](#next-steps) [Prevent documentation drift ](/agent/prevent-documentation-drift/)Set up automatic drift detection and a weekly scheduled audit [Automatic PR review ](/agent/automatic-pr-review/)Understand the confidence tiers that decide when EkLine acts [Update and review ](/agent/update-review/)Update docs on demand from a ticket or a code change [Automations ](/agent/scheduled-agents/)Automate recurring documentation maintenance tasks # Keep your team informed > Send a weekly report of EkLine's docs work and a brief of what shipped to Slack, using the EkLine Work Report and Release Marketing Brief automations. Two built-in automations post to Slack each week: a report of the docs work EkLine did, and a brief of the customer-facing changes that shipped. ## What each one produces [Section titled “What each one produces”](#what-each-one-produces) | | **EkLine Work Report** | **Release Marketing Brief** | | ------------------ | ----------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- | | For | Docs leads, and whoever reviews EkLine’s pull requests | Marketing and customer success teams | | Covers | Pull requests EkLine opened in your docs repositories in the time window. Default: last 7 days. Maximum: 3 months. | Customer-facing changes that shipped in the lookback window. Default: last 7 days. | | Contains | Counts at a glance, then **Delivered**, **Awaiting your review** (with who each pull request is waiting on), **Set aside**, and **Notes** | Up to three changes, each with a headline, what it does, why it matters, and a docs link when the agent finds one, plus a count of smaller improvements | | Leaves out | Pull requests opened by anyone other than EkLine | Reverted changes, security fixes, work behind a feature flag that is not fully rolled out, internal changes, and docs-only changes | | Quiet week message | “EkLine had no documentation activity in this period.” | “No significant customer-facing changes shipped in this period.” | The agent writes each post, so the layout can vary from the one above. The work report runs as the Docs Agent and the brief as the Marketing Agent. Both are read-only: they do not open, change or close any pull request. In the work report, the agent @-mentions each reviewer in Slack when it can match their GitHub or GitLab account to a Slack user. It tries a mapping you give first, then email, then full name. Reviewers it cannot match are listed under **Notes** and are not notified. For the release brief, the agent reads whatever it can reach: merged pull requests, closed tickets, changelog files and release tags. ## Where results go [Section titled “Where results go”](#where-results-go) The agent posts the result to the channel you set under **Send a Slack notification**. Every result is also in the run’s session. If no post appears, open the session to see why. If Slack is not connected, the agent says so in the session and does not post. EkLine does not send a message when an automation runs, succeeds or fails, in Slack or by email. A run that fails or is skipped posts nothing. Check a run’s status in the automation’s **Runs** tab. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * Slack is connected: in **Settings** > **Integrations**, click **Connect** on the **Slack** card. See [Slack](/agent/slack-bot/). * For the work report, your docs repositories are selected under **Where EkLine writes**. See [Connect GitHub](/agent/github-app-setup/) or [Connect GitLab](/agent/gitlab-setup/). * For the brief, your organization has the Marketing Agent. If you don’t see the **Marketing** tab under **Automations**, contact . ## Steps [Section titled “Steps”](#steps) 1. Click **Automations** in the sidebar, then the **Releases and reports** tab. 2. On the **EkLine Work Report** card, click **Set up**. 3. Keep **Time window** at `last 7 days`, or change it, for example to `last 2 weeks`. 4. Under **Schedule**, keep the default of every week on Monday at 9:00 AM, or select another time. 5. Under **Notifications**, keep **Send a Slack notification** checked and enter the channel name. Leave **Raise a pull request** unchecked: the report changes nothing. 6. To make sure a reviewer is @-mentioned, add a mapping under **Additional instructions (optional)**, for example `Reviewer Slack IDs: jdoe → U0123ABCD`. 7. Click **Turn on**. 8. Repeat for **Release Marketing Brief** under the **Marketing** tab, with **Lookback window** and a channel your marketing team reads. The brief runs as the Marketing Agent, so its panel has no **Raise a pull request** option. To get either one once, outside its schedule, enter `/ekline-work-report` in a Docs Agent session or `/ekline-release-marketing-brief` in a Marketing Agent session, with the window and the channel, for example `/ekline-work-report last 2 weeks, post to #docs-team`. ## Check it worked [Section titled “Check it worked”](#check-it-worked) 1. Click **Edit** on the automation’s card, open the **Runs** tab, and click **Run now**. 2. When the run shows **Succeeded**, click its start time. The session shows the report and whether the Slack post went through. 3. Check the channel for the post. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Fix | | ------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The session says Slack is not connected and it skipped the post | Connect Slack in **Settings** > **Integrations**, then click **Run now**. | | The session says the post failed and the bot may not be a member of the channel | Add the EkLine app to the channel in Slack, then click **Run now**. | | A reviewer is listed under **Notes** instead of being @-mentioned | Add a mapping for their username under **Additional instructions (optional)**, as in step 6. | | The work report refuses the time window | It covers at most 3 months. Enter a shorter window. | | The run shows **Skipped** | The reason is under the run’s start time: your organization was out of credits (see [Credits and usage](/workspace/credits-and-usage/)), or, for the brief, the Marketing Agent is not enabled for your organization. | ## Next [Section titled “Next”](#next) * [Automations](/agent/scheduled-agents/): schedule, edit and turn off automations. * [Automation templates](/reference/automation-templates/): every built-in automation, including **Release Notes Generator**, which drafts release notes as a pull request. * [Slack](/agent/slack-bot/): ask EkLine for docs changes from a Slack thread. # Manage a Confluence or Pylon knowledge base > Use Docs Agent to keep a Confluence or Pylon knowledge base accurate: update pages across a space, improve articles, and publish changes back. Docs Agent helps internal teams keep a knowledge base accurate and up to date. Connect an internal Confluence space, such as an engineering or support team’s knowledge base, or a Pylon knowledge base used by your support team. The agent updates individual pages, maintains an entire space, and improves articles from sources like support tickets, then publishes the changes straight back to the source with no copying between tools. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * A connected Confluence or Pylon knowledge base with at least one space or knowledge base marked as managed (see [Connect your knowledge base](#connect-your-knowledge-base)). * Permission to edit the page or article you want to update. ## Connect your knowledge base [Section titled “Connect your knowledge base”](#connect-your-knowledge-base) An organization administrator connects the knowledge base once. The agent can then read from and write to the spaces or knowledge bases you select. If you don’t see **KB Management** under **Settings** > **Integrations**, contact . * Confluence 1. Connect Confluence under **Settings** > **Integrations**. 2. Go to **Settings** > **Integrations** > **KB Management**. 3. Enable **Manage Confluence**. 4. Select each Confluence space you want the agent to manage. Your changes save automatically. A **Saving…** indicator appears while each change saves, then changes to **Saved**. The agent can now read and update pages in the selected spaces, including [managing images](#manage-images-in-a-confluence-page) on them. * Pylon 1. Go to **Settings** > **Integrations** > **KB Management**. 2. Enable **Manage Pylon**. 3. Paste your **Pylon API token** and click **Connect**. 4. Select each Pylon knowledge base you want the agent to manage. 5. Set a **Default author**. New articles the agent creates are attributed to this Pylon user. Your changes save automatically. A **Saving…** indicator appears while each change saves, then changes to **Saved**. The agent can now read, update, and create articles in the selected knowledge bases. ## Step 1: Open the editor [Section titled “Step 1: Open the editor”](#step-1-open-the-editor) 1. Sign in to the [EkLine app](https://ekline.io/dashboard). 2. Click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. ## Step 2: Retrieve the page [Section titled “Step 2: Retrieve the page”](#step-2-retrieve-the-page) Ask the agent to fetch the page you want to update. The agent loads it into the editor panel as Markdown you can edit. * Confluence Paste the page URL: ```text Retrieve this Confluence page and fix the outdated CLI flags in the installation section: https://your-org.atlassian.net/wiki/spaces/ENG/pages/123456 ``` * Pylon Name the article, or list the knowledge base first: ```text List the articles in our Pylon knowledge base, then retrieve the "Resetting your password" article so I can update it. ``` ## Step 3: Review the retrieved content [Section titled “Step 3: Review the retrieved content”](#step-3-review-the-retrieved-content) Read through the page in the editor and confirm it matches what you expect to see in the knowledge base. Editing a page you retrieved keeps your changes scoped to that page. ## Step 4: Edit with follow-up prompts [Section titled “Step 4: Edit with follow-up prompts”](#step-4-edit-with-follow-up-prompts) Refine the page by continuing the conversation. The agent keeps the context from your first prompt, so you can build on the content incrementally. You can also edit directly in the editor panel. * Update a section ```text Rewrite the "Requirements" section to list the new minimum version, and add a note about the deprecated config option. ``` * Improve clarity ```text Tighten the troubleshooting steps into a numbered list and remove the duplicated paragraph about cache clearing. ``` * Add from a source ```text Add a new FAQ entry based on this Slack thread: https://workspace.slack.com/archives/C01234/p1234567890 ``` ## Step 5: Publish your changes [Section titled “Step 5: Publish your changes”](#step-5-publish-your-changes) When the page is ready, publish it back to the source. 1. Enable **View All Changes** in the toolbar to review a diff of your edits. 2. Make any final adjustments in the editor panel. 3. Click **Update KB**. The chat box fills with `Publish my knowledge base edits`. Press **Enter** to send it, and the agent publishes your edits. The **Update KB** button replaces **Raise PR** when you work on a knowledge base, because there is no pull request to open. Your approval publishes the edits to the source instead. Caution Publishing overwrites the live page immediately. Your edited content replaces Confluence pages, and Pylon articles update in place. Review the diff before you publish. For Pylon, publishing does not change whether an article is listed as published or unpublished. ## Verify the update [Section titled “Verify the update”](#verify-the-update) Confirm the change reached the source: * **Open the live page.** The Confluence page or Pylon article shows your edits. * **Check the structure.** Headings, tables, and links render the way they did before, with your changes applied. * **Confirm the diff matched.** What you reviewed under **View All Changes** is what appears on the live page. ## Manage images in a Confluence page [Section titled “Manage images in a Confluence page”](#manage-images-in-a-confluence-page) On a Confluence page, the agent works with images as well as text. It can add a new image, replace an existing one, or remove an image. It displays each image inline in the editor as you work, so you see the page the way it renders in Confluence. The agent adds any image it can produce or find during a session. For example, ask it to [capture a screenshot from a video](/agent/screenshots-from-video/) or from [browsing an authenticated page](/agent/sandbox/), then place that screenshot on the page: ```text Retrieve this Confluence page, then browse to our dashboard, take a screenshot of the Settings panel, and add it under the "Settings" heading: https://your-org.atlassian.net/wiki/spaces/ENG/pages/123456 ``` To swap or clean up existing images, describe the change: * Replace an image ```text Replace the outdated dashboard screenshot with a fresh capture of the current dashboard. ``` * Remove an image ```text Remove the second screenshot in the "Overview" section. It no longer matches the UI. ``` Images change on the live page only when you publish. When you publish with **Update KB**, the agent uploads the images you added as attachments before it updates the page, and deletes the attachments for any images you removed. If you close the editor or start a different task without publishing, the Confluence page and its attachments stay untouched. ## Maintain the whole knowledge base [Section titled “Maintain the whole knowledge base”](#maintain-the-whole-knowledge-base) The same workflow scales beyond a single page. Point the agent at a managed space or knowledge base, and it can review many pages at once, apply consistent updates, and, in Pylon, create new articles. You publish each change back with **Update KB**. * Audit a space ```text Review the Engineering space for pages that still reference the old deployment command. Update each one to the new command and summarize what you changed. ``` * Improve from support tickets ```text Our support team keeps answering the same question about SSO setup. Draft a new help center article that covers it, based on this thread: https://workspace.slack.com/archives/C01234/p1234567890 ``` * Keep articles current ```text Compare the "Billing FAQ" article against the latest pricing page and update any answers that are out of date. ``` Review each change under **View All Changes** before you publish. For new Pylon articles, the agent uses the default author you set in **KB Management**. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause | Fix | | -------------------------------------------------------- | -------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------ | | The space or knowledge base isn’t available to the agent | It isn’t marked as managed, or the integration isn’t connected | Enable it under **Settings** > **Integrations** > **KB Management** and confirm you selected the space or knowledge base | | The agent can’t retrieve a page | The connected account lacks permission, or the URL is wrong | Verify you can open the page yourself, then paste the exact page URL | | The **Update KB** button is disabled | There are no changes to publish | Make an edit first. The button is enabled once the page differs from the retrieved version | | A new Pylon article can’t be created | No default author is set | Set a **Default author** under **Settings** > **Integrations** > **KB Management** | | Publishing a Confluence page with an image fails | Confluence was connected before image management was available | Reconnect Confluence under **Settings** > **Integrations**, then publish again | ## Next steps [Section titled “Next steps”](#next-steps) * [Update and review](/agent/update-review/): keep documentation in sync with code changes. * [Integrations](/agent/integrations/): reference Slack, Notion, Linear, and more in your prompts. * [Customize for your organization](/agent/custom-instructions/): match your knowledge base’s tone and structure with custom instructions. # Connect an MCP client > Start and monitor EkLine agent sessions from Claude Code, Claude on the web, Cursor, or another MCP client that can send a request header, using a personal API key. EkLine serves a [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server, so you can work with EkLine from your own AI client without switching to the web app. Describe what you want to your client (Claude Code, Claude on the web, Cursor, or another) and it calls EkLine for you: start an agent session, find past work, ask how a feature works, or set your organization up. For every tool and its parameters, see [MCP tools](/reference/mcp-tools/). ## Prerequisites [Section titled “Prerequisites”](#prerequisites) * You are a member of an EkLine organization. * You use an MCP client that connects over HTTP and lets you set a request header. Claude Code, Claude on the web, Claude Desktop and Cursor all qualify. See [Supported clients](#supported-clients). ## Create a personal API key [Section titled “Create a personal API key”](#create-a-personal-api-key) Your client authenticates with a personal API key. The key acts as you, in one organization: every session you start through it is recorded under your name, and no other member can see or use it. 1. Open **Settings** > **Profile** > **Personal API keys**, or go directly to [your API keys](https://ekline.io/settings/profile/api-keys). 2. Confirm the organization named on the page is the one you want the key to act in. A key works in a single organization. To create a key for another organization, switch to it first. 3. Click **Create key**. Under **What is it for?**, enter a name you will recognize later, such as `Claude Code on my laptop`, and click **Create key**. 4. The **Set up your client** dialog shows the key once, inside a setup snippet for each client. Click your client’s tab, click **Copy setup**, and keep the snippet somewhere safe, such as a password manager. Then click **Done**. Caution The key is shown once, at creation, and never again. If you lose it, revoke it and create a new one. A key looks like `ek__`. The `` half is not secret (you see it in your key list long after the secret is gone), but the whole value is a credential. Treat it like a password and never commit it to a repository. ## Add EkLine to your client [Section titled “Add EkLine to your client”](#add-ekline-to-your-client) The **Set up your client** dialog gives you a setup snippet for each client with your key filled in. After you close it, the **Connecting a client** section of the API keys page shows the same snippets with `YOUR_KEY` where the key goes. Use the snippet for your client and follow the instructions below. The server URL is `https://ekline.io/api/mcp`. * Claude Code Run this in your terminal, then open a new session, since Claude Code loads MCP servers at start: ```bash claude mcp add --transport http ekline \ https://ekline.io/api/mcp \ --header "Authorization: Bearer ek__" ``` * Cursor Add this to `.cursor/mcp.json`, in your home folder or your project: .cursor/mcp.json ```json { "mcpServers": { "ekline": { "url": "https://ekline.io/api/mcp", "headers": { "Authorization": "Bearer ek__" } } } } ``` * Other clients Any client that lets you set a request header works, including VS Code and Windsurf. Point it at the server URL and send your key as a bearer token: ```text Server URL https://ekline.io/api/mcp Header Authorization: Bearer ek__ ``` Once the client reconnects, it lists the EkLine tools. That confirms your key resolved and the server is reachable. ### Connect from Claude on the web or Claude Desktop [Section titled “Connect from Claude on the web or Claude Desktop”](#connect-from-claude-on-the-web-or-claude-desktop) Claude on the web and Claude Desktop do not take the command line setup. Add EkLine as a custom connector instead, with your key sent as a request header. Both use the same connector dialog. Request headers is not on every plan Claude shows **Request headers** only for some organizations, as a beta. If you do not see it, your plan cannot use an API key yet. On Team and Enterprise plans, the Owner adds the connector and every member uses the Owner’s key, so EkLine records every change made through it under the Owner’s name. 1. Open **Settings** > **Connectors** and click **Add** > **Add custom connector**. 2. Give the connector a name, such as `ekline`, enter the server URL `https://ekline.io/api/mcp`, and click **Continue**. ![The Add custom connector dialog in Claude settings, with the connector named ekline and the server URL https://ekline.io/api/mcp entered](/assets/images/claude-add-custom-connector.png) 3. Under **Authentication**, select **No sign-in**. Claude detects this for EkLine, which authenticates with a key rather than a sign-in flow. 4. Under **Request headers**, add a header named `Authorization` with the value `Bearer ek__`, then click **Connect**. ![The custom connector authentication panel in Claude with No sign-in selected and an Authorization request header set to a Bearer token](/assets/images/claude-connector-no-signin-header.png) The EkLine tools then appear under the connector’s **Tool permissions**, which confirms your key resolved. ## Start a session [Section titled “Start a session”](#start-a-session) Ask your client to use the `create_session` tool, and describe the task in your own words, including any references you have, such as a pull request, a ticket, or a doc path. For example: > Use EkLine to document the retry behavior added in PR #4821. The reasoning is in the linked ticket. The tool takes two inputs: | Input | Required | Description | | -------- | -------- | ------------------------------------------------------------------------------------------------------------------------ | | `prompt` | Yes | The task, in your own words, including the reasoning behind it. | | `agent` | No | Which agent runs the session: `docs` (the default) for documentation work, or `marketing` for a Marketing Agent session. | `create_session` returns a session id and a link to follow the work in the app. The session appears in EkLine with you as its author. ## Check how a session is going [Section titled “Check how a session is going”](#check-how-a-session-is-going) Ask your client to use the `get_session_status` tool with the session id that `create_session` returned. It reports the current status and the session’s most recent messages, so your client can tell you what the agent is working on rather than only that it is busy. The status is one of: | Status | Meaning | | ----------- | -------------------------------------------------------------------------------------------------- | | `working` | The agent is still running the session. | | `completed` | The session finished successfully. | | `failed` | The session ran into a problem, or has had no activity for more than 15 minutes without finishing. | | `cancelled` | The session was stopped. | Ask for the status whenever you want an update, including while the session is still running. A session keeps running to completion even after `create_session` returned, so checking in does not interrupt it. To open the session yourself, follow the link in either tool’s response. Docs sessions open in the editor; marketing sessions open in the Marketing Agent. ## Redirect a running session [Section titled “Redirect a running session”](#redirect-a-running-session) A session does not end when `create_session` returns. To correct a running session, add a requirement, or hand it a reference you left out, ask your client to use the `send_session_message` tool with the session id and your message: > Tell session abc123 to also cover the retry limits, which are in the linked ticket. The tool takes two inputs: | Input | Required | Description | | ----------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | `sessionId` | Yes | The session to send the message to. Use the id that `create_session` returned, or copy it from your [session list](/agent/find-sessions/). | | `prompt` | Yes | What to tell the agent, in your own words, including the reasoning behind it. | The message joins the session you named rather than starting a new one. If the session id is wrong, the tool reports that the session was not found instead of creating one. ## Find your sessions [Section titled “Find your sessions”](#find-your-sessions) Ask your client to use the `list_sessions` tool to see your organization’s recent sessions: what each was asked to do and when it was last active. Many sessions list no pull request even when they opened one. To check, call `get_session_details` with the session id, or open the session. Use it before you start a new session, to check whether the work is already under way. Paste a pull request, merge request, Jira, or session link and the tool returns the sessions related to it, or describe what you are looking for in plain words. `list_sessions` does not report whether a session is still running. It shows when a session was last active and whether someone marked it done. To read a session’s live status, call `get_session_status` with its id. ## Read a session in full [Section titled “Read a session in full”](#read-a-session-in-full) Ask your client to use the `get_session_details` tool with a session id. It returns what the session was asked, where it came from (the editor, Slack, or a scheduled job), the pull requests it opened, and its conversation, oldest first. Each pull request shows its number, title, state, link, and the branch it went into. The state is the one EkLine last recorded, and the answer says when that was. Use it with `list_sessions` to check what earlier work did before you repeat it. For example, before you ask for a guide, check whether an earlier run of a scheduled job already wrote it. A long conversation comes back with messages from the middle left out. The request and the last report are always included, and the answer says how many messages were left out. To read the rest, open the session. To follow a session that is still running, use `get_session_status` instead. ## Read a report from a link [Section titled “Read a report from a link”](#read-a-report-from-a-link) Paste a link to a journey, one of its reports or runs, a [collection](/journeys/collections/) or one of its reports, or a shared report into your client and ask what it says. Your client reads it with the tool that matches the link, such as `get_journey_report` or `get_collection_report`. For example: > What does this report say about how often we’re recommended, and what should we fix first? The tools read reports from your own organization only. A link from another organization, or a shared link that was turned off, reads as not found. Reading a report doesn’t change it. A collection run that’s still going shows how many answers are in, and its written summary appears after the run finishes. If your client doesn’t list these tools, contact . See [MCP tools](/reference/mcp-tools/#journeys-tools) for every journeys and collections tool. ## Create, run and share a collection [Section titled “Create, run and share a collection”](#create-run-and-share-a-collection) Ask your client to group published journeys into a [collection](/journeys/collections/) with `create_collection`, rename it or change its journeys with `update_collection`, and run it with `run_collection`. A run sends each journey’s questions to the AI agents it names and uses credits, so ask for one only when you want it. When it finishes, read the report with `get_collection_report`. To share a finished report with someone outside EkLine, ask for a link: `share_collection_report` returns the report’s link, and anyone with it can open the report without signing in. To turn a link off, open the report in EkLine and click **Share**, then **Stop sharing**. EkLine’s own agents can create, change and run collections, but only you can make a share link. The Buyer questions collection’s journeys can’t be changed. To delete a collection, use EkLine. ## Add reference documents [Section titled “Add reference documents”](#add-reference-documents) Ask your client to add a [reference document](/agent/reference-documents/) from text with `add_knowledge_base_document`, such as a product brief or a glossary in Markdown, and to list your reference documents with `list_knowledge_base_documents`. A document added this way is processed like an uploaded file: EkLine gives it a title and a summary. Both agents use it in sessions that start after it’s ready, and the list shows whether each document is ready. The tool refuses text that’s already a reference document. To add a file that isn’t text, such as a PDF or a Word document, or to delete a document, use **Settings** > **Organization** > **Knowledge Base**. EkLine’s own agents can list your reference documents, but only you can add one. If your client doesn’t list these tools, contact . ## Ask about EkLine [Section titled “Ask about EkLine”](#ask-about-ekline) Ask your client to use the `search_ekline_docs` tool with a question about EkLine, such as how a feature works or what a reviewer rule means. It returns the relevant sections of the product documentation and a link to each page, so your client answers from the docs rather than from memory. ## Set up your organization [Section titled “Set up your organization”](#set-up-your-organization) Two tools write settings that apply to your whole organization, not only the current session. Sessions that start after the change use them. Caution `update_company_profile` and `create_rules` change organization-wide settings and take effect immediately, with no review step. Ask for a change only when you intend every member’s sessions to use it. ### Set the company profile [Section titled “Set the company profile”](#set-the-company-profile) Ask your client to use the `update_company_profile` tool to set your organization’s [company profile](/agent/company-profile/). The Marketing Agent reads it in every session, and the Docs Agent reads it when company context is turned on for your organization. The tool sets these fields: | Field | What it describes | | ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | | `companyProfile` | What your company does. | | `domain` | Your company’s website, such as `acme.com`, without `https://` or `www.`. Journeys reports use it to recognize your pages. Agents don’t read it. | | `keyFeatures` | The product capabilities worth naming. | | `audience` | Who your documentation is written for. | | `competitors` | Who your company is compared against, in your own words. The named competitor list is changed as described in the next section. | | `toneAndVoice` | How your documentation should sound. | | `visualStyle` | How images the agents make should look: your brand colors, one per line as a role and a hex code (`Primary: #cb2e2f`), then any notes on style. | | `additionalContext` | Anything else the agent should know. | Each field you set replaces its current value in full. Fields you leave out keep their current value, and the tool reports what each changed field held before, so you can confirm the result. To clear a field, use EkLine settings. Because each field you set replaces the old value, ask your client to read the profile with `get_company_profile` first. See and change the company profile in **Settings** > **Organization** > **Company Profile**. If you don’t see it, contact . ### Change brand names, competitors, personas and documentation sites [Section titled “Change brand names, competitors, personas and documentation sites”](#change-brand-names-competitors-personas-and-documentation-sites) Ask your client to use the `update_company_profile` tool to add or change rows in your company profile’s **Brand names**, **Competitor profiles**, **Personas** and **Documentation sites**. Name the row you want changed as it appears in your profile. For example: > Rename the brand name Acme Cloud to Acme Platform, add Globex and Initech as competitors, and rewrite the Support Lead persona to say they answer tickets but don’t change billing. The tool makes every change you ask for in one request or none of them, including changes to the fields above. If one change can’t be made, such as a persona name that doesn’t exist or a competitor that’s already listed, it changes nothing and says which change failed and why. When the changes are made, it reports each row as it was and as it is now. The agents read these rows with the rest of your company profile, and [Journeys](/journeys/) reports use your brand names and competitor profiles to find your company and its competitors in AI answers. To remove a row or clear a field, use EkLine settings. See [MCP tools](/reference/mcp-tools/#named-entity-changes) for the fields each row takes. ### Add a rule [Section titled “Add a rule”](#add-a-rule) Ask your client to use the `create_rules` tool to add a standing instruction the agents follow. Give it a name, the instruction itself, and the type of rule: * A **general** rule (the default) applies to every task the agent works on. * A **review** rule is applied to every documentation change the agent makes before it opens or updates a pull request. You can also pass file globs, such as `docs/**/*.mdx`, to limit a rule to matching files. Omit them to apply the rule everywhere. `create_rules` only adds rules. If you ask for a rule whose name your organization already uses, the tool refuses and names the existing rule rather than creating a duplicate. To change, disable, or delete a rule, use EkLine settings. Manage rules in **Settings** > **Docs Agent** > **Rules**. ## How keys and attribution work [Section titled “How keys and attribution work”](#how-keys-and-attribution-work) * **Keys are personal.** A session you start through your key is recorded under your name in the app, the same as a session you start in the web app. * **Membership is checked on every call.** If you leave an organization, your keys for it stop working on the next request, and you do not need to revoke them first. * **Revoking takes effect immediately.** To revoke a key, open your **Personal API keys** page and click **Revoke** next to it. The key stays in your list, marked revoked, and stops authenticating on its next use. * **Your key is never forwarded.** EkLine does not pass it to the agent. ## Supported clients [Section titled “Supported clients”](#supported-clients) | Client | Supported | | ------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- | | Claude Code | Yes | | Claude on the web | Yes, as a custom connector. See [Connect from Claude on the web or Claude Desktop](#connect-from-claude-on-the-web-or-claude-desktop). | | Claude Desktop | Yes, as a custom connector. See [Connect from Claude on the web or Claude Desktop](#connect-from-claude-on-the-web-or-claude-desktop). | | Cursor | Yes | | ChatGPT | Not yet. ChatGPT connectors sign in with OAuth 2.1 and do not accept an API key. | | Other clients that let you set a request header (VS Code, Windsurf, and others) | Yes | ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Symptom | Cause and fix | | ------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The client lists no EkLine tools | The key did not resolve. Check that the `Authorization` header reads `Bearer` followed by your full key, that the key is not revoked, and that you are still a member of the organization it was created for. A revoked, removed, or unknown key all produce an empty tool list. | | The `claude mcp add` command does not work in a browser | That command sets up the Claude Code command line. In Claude on the web or Claude Desktop, add EkLine as a custom connector instead. See [Connect from Claude on the web or Claude Desktop](#connect-from-claude-on-the-web-or-claude-desktop). | | `create_session` returns an error about reaching EkLine | EkLine could not be reached and no session was started. Try again. | | `create_session` reports a session was accepted but its id did not arrive | The session is running but its id did not return in time. Find it in your [session list](https://ekline.io/) rather than starting another one. | | `get_session_status` says it could not read the session | EkLine could not be reached to read the session. The session may still be running, so try again rather than starting a new one. | | `get_session_details` says it could not read the session’s pull requests | EkLine could not read them this time. It does not mean the session opened none. Try again before you rely on the answer. | | `send_session_message` says the session was not found | The session id is wrong, or the session belongs to another organization. Check the id in your [session list](/agent/find-sessions/). | | `send_session_message` returns an error about reaching EkLine | EkLine could not be reached and the message was not sent. Try again. | ## Next steps [Section titled “Next steps”](#next-steps) * [Find a past session](/agent/find-sessions/). Search your history or paste a link to reopen any session. * [Getting started with Docs Agent](/agent/getting-started/). Create your first document in the editor. * [Update and review](/agent/update-review/). Resume a session to keep docs in sync with code changes. # Organization memory > How EkLine's agents keep notes between sessions for your organization, who writes them, and how they differ from custom instructions. Organization memory is the notes EkLine’s agents keep between sessions. A later session can start from what earlier ones learned about your organization. ## Who writes it [Section titled “Who writes it”](#who-writes-it) The agent writes it. While it works, the agent saves notes it expects to need again, and later sessions read them. You don’t write or approve these notes. No one has to turn memory on. ## What it is shared across [Section titled “What it is shared across”](#what-it-is-shared-across) Your organization has one memory. Every session reads and writes it: * Both agents: Docs Agent sessions and Marketing Agent sessions use the same memory. * Every way a session starts: the editor, a GitHub pull request, a GitLab merge request, Slack, Discord, a Jira ticket, an automation or a Journeys report. * Every member: a note saved in your session can shape a teammate’s session later. Memory is kept on EkLine’s servers. Each organization has its own, and a session uses only its own organization’s memory. For how EkLine handles your data, see the [Trust Center](https://ekline.io/trust-center). ## What you can and can’t do with it [Section titled “What you can and can’t do with it”](#what-you-can-and-cant-do-with-it) You can’t view or edit organization memory in the app today. There is no page for it in **Settings**, and it is not shown in the editor. If an agent acts on something wrong or out of date, put the correct rule in [Custom instructions](/agent/custom-instructions/). The agent follows them in every new session. ## How it differs from Custom instructions [Section titled “How it differs from Custom instructions”](#how-it-differs-from-custom-instructions) | | Organization memory | Custom instructions | | ----------------------- | ---------------------------------------- | ---------------------------------- | | Who writes it | The agent | You | | Can you see and edit it | No | Yes, under **Settings** | | Which agents | Both agents share one memory | Each agent has its own | | Best for | Details the agent picks up while working | Rules you want followed every time | Use [Custom instructions](/agent/custom-instructions/), [Company Profile](/agent/company-profile/) and [repository notes](/agent/reference-documents/) for anything you want to control. Treat memory as the agent’s own notes, not as a setting. ## FAQ [Section titled “FAQ”](#faq) ### Why does the agent remember something from another session? [Section titled “Why does the agent remember something from another session?”](#why-does-the-agent-remember-something-from-another-session) Every session in your organization shares one memory, across both agents and every member. A note one session saved can appear in another. ### Can I see what is in memory? [Section titled “Can I see what is in memory?”](#can-i-see-what-is-in-memory) No. The app has no view of organization memory today. ### Can I delete or correct a memory? [Section titled “Can I delete or correct a memory?”](#can-i-delete-or-correct-a-memory) Not directly. Put the correct rule in [Custom instructions](/agent/custom-instructions/) for the agent that got it wrong. ### Does memory replace Custom instructions? [Section titled “Does memory replace Custom instructions?”](#does-memory-replace-custom-instructions) No. Memory is what the agent noted for itself. Custom instructions are what you tell it, and you can read and change them at any time. ### Is memory shared with other organizations? [Section titled “Is memory shared with other organizations?”](#is-memory-shared-with-other-organizations) No. Each organization has its own memory. # Prevent documentation drift > Set up automatic drift detection and a weekly scheduled review so your documentation stays in sync as code changes. Documentation drift is the gap that opens each time a PR merges without a matching docs update. Two safeguards catch it: | Safeguard | When it runs | What it does | | ------------------------- | -------------------- | ------------------------------------------------------------------------------------- | | **Automatic PR review** | When a PR opens | Analyzes the diff and triggers a docs update when EkLine detects documentation impact | | **Scheduled drift audit** | On a cadence you set | Reviews recent code activity and fixes the highest-priority undocumented change | ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * Automatic PR review, which is on by default, and Automations. If you don’t see **Automations** in the sidebar, contact . * The [EkLine GitHub App connected](/agent/github-app-setup/), with your docs repository selected under **Where EkLine writes** and the repository where code changes land selected under **What else EkLine can read**. ## Step 1: Verify automatic PR review is enabled [Section titled “Step 1: Verify automatic PR review is enabled”](#step-1-verify-automatic-pr-review-is-enabled) Automatic PR review runs only on repositories that are selected and monitored. Confirm monitoring is on for the repository where code changes land. 1. Go to your [EkLine dashboard](https://ekline.io/dashboard). 2. Go to **Settings** > **Integrations** > **GitHub Integration**. 3. Find the repository where code changes land, under **What else EkLine can read** (or **Where EkLine writes**), and confirm its **Enable monitoring** toggle is on. 4. If the toggle is off, enable it. EkLine saves the change immediately. By default, EkLine reviews pull requests into the repository’s default branch. It skips drafts and pull requests from common bots. It posts a comment when it finds docs impact. To change which pull requests it reviews, see [review rules](/agent/review-rules/). ## Step 2: See it detect a real change [Section titled “Step 2: See it detect a real change”](#step-2-see-it-detect-a-real-change) To confirm automatic PR review works, open a pull request with a code change that affects your docs. 1. Open a pull request against a monitored repository. Use a change that clearly affects your docs, such as adding a new API endpoint, renaming a configuration option, or removing a deprecated feature. 2. Make sure the PR is **ready for review** and not a draft. 3. Wait about 30 seconds. EkLine reads the diff, title, and description. Then it compares the changes with your docs. 4. Check the PR comments. EkLine posts one of the following: | Result | What it means | | -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | No comment | Low confidence: the change likely does not affect documentation | | Suggestion comment | Medium confidence: EkLine suggests a docs update and mentions `@ekline-ai` so you can start it yourself | | Auto-trigger comment | High confidence: the comment starts the Docs Agent on the update. EkLine replies with a link to the session, and the agent opens a docs PR. | If EkLine posts a comment, automatic drift detection is working. ## Step 3: Set up a weekly scheduled drift audit [Section titled “Step 3: Set up a weekly scheduled drift audit”](#step-3-set-up-a-weekly-scheduled-drift-audit) Automatic PR review catches impact as PRs open. The drift audit catches what it missed: changes that merged before you connected EkLine, and changes it did not rate High. 1. Click **Automations** in the sidebar. 2. Find **Docs Drift Review** under the **Accuracy** tab, or enter its name in **Search**. 3. Click **Set up** on its card. 4. In **Lookback window**, enter how far back each run looks, for example `last 30 days`, or `last 7 days` for a tighter weekly cadence. The field starts empty, and the automation cannot turn on until you fill it in. 5. Under **Schedule**, keep the default of every week on Monday at 9:00 AM, or change **Repeat every**, **Repeat on** and **At**. Monday morning is a common choice. It sweeps last week’s changes before your team starts new work. 6. Under **Notifications**, keep **Raise a pull request** checked, so the agent opens a PR with the docs fix. 7. **Send a Slack notification** is also checked. Enter your docs channel name, for example `docs-team`, and the agent posts a run summary there each week. Or clear the checkbox. 8. Click **Turn on**. The **Docs Drift Review** card now shows its toggle on, and the automation runs at the next scheduled time. ## Verify the setup [Section titled “Verify the setup”](#verify-the-setup) Confirm both safeguards are active: * [ ] **Enable monitoring** is on for the repository where code changes land, in **Settings** > **Integrations** > **GitHub Integration**. * [ ] The **Docs Drift Review** card on the **Automations** page has its toggle on. To see the automation’s output, click **Edit** on its card and open the **Runs** tab. Under **Recent runs**, click a run’s start time to open the full Docs Agent session. To get a run before the first scheduled time, click **Run now** on the same tab. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | ------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | EkLine never comments on PRs | Check that the GitHub App is installed on the repository. Check that the repository is selected under **Where EkLine writes** or **What else EkLine can read**, and that **Enable monitoring** is on. EkLine skips draft PRs by default, so mark them ready for review. A Low assessment posts no comment. | | Auto-trigger fired but no docs PR appeared | The agent may still be working. Wait up to a minute, then open the session from the link in EkLine’s reply on the PR. If **Auto-create documentation PRs** is off in the repository’s review rules, the drafts wait in the editor instead. | | An automation run shows **Failed** | On the automation’s **Runs** tab, click the run’s start time to open the full session. Common causes: the repository is no longer accessible, or there were no code changes in the audit window. | | Docs PR from drift review is off-target | Comment on the docs PR and mention `@ekline-ai` with feedback. The agent resumes the session and pushes corrections. | ## Next steps [Section titled “Next steps”](#next-steps) * [Automatic PR review](/agent/automatic-pr-review/): understand the confidence tiers and what each detection level means. * [Automations](/agent/scheduled-agents/): explore the full template library for recurring documentation tasks. * [GitHub PR bot](/agent/github-integration/): trigger documentation sessions manually from a PR comment. * [Create documentation](/agent/create/): generate READMEs, API references, and guides from your codebase. # Reference documents and repository notes > Upload reference documents and describe each repository so the Docs Agent and the Marketing Agent have context your code does not hold. Reference documents and repository notes give EkLine’s agents context that is not in your code: files you upload, and a short note on what each repository holds. | Context | Where you add it | Which agents read it | | ------------------- | ---------------------------------------------------------------------------------- | ------------------------------ | | Reference documents | **Settings** > **Organization** > **Knowledge Base** | Docs Agent and Marketing Agent | | Repository notes | **Settings** > **Integrations** > **GitHub Integration** or **GitLab Integration** | Docs Agent and Marketing Agent | ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * The **Knowledge Base** tab under **Settings** > **Organization**, for reference documents * A repository saved on your GitHub or GitLab integration, for repository notes Any member of your organization can upload and manage reference documents and edit repository notes. ## Upload a reference document [Section titled “Upload a reference document”](#upload-a-reference-document) 1. Log in to your [EkLine dashboard](https://ekline.io/dashboard). 2. Go to **Settings** > **Organization** > **Knowledge Base**. 3. In the **Upload Document** section, drag a file onto the upload area, or click it to select a file. You can upload PDF, Word, PowerPoint, Excel, Markdown, text, HTML, CSV, AsciiDoc, reStructuredText, image and SVG files. Each file can be up to 3 MB. Upload one file at a time. 4. Wait for the upload to finish. EkLine processes the document and adds it to the **Uploaded Documents** list with a generated title and a one-line summary. Markdown, text, HTML, CSV, AsciiDoc and reStructuredText files are always processed. PDF, Office, image and SVG files are always stored, but EkLine may not be able to read them yet. See [Troubleshooting](#troubleshooting). Both agents use your documents in every session started after the upload finishes. You can start the session in the editor or trigger it from a [GitHub pull request](/agent/github-integration/), [GitLab merge request](/agent/gitlab-integration/), or [Slack](/agent/slack-bot/). ## What to upload [Section titled “What to upload”](#what-to-upload) Upload context the agent can’t get from reading your code or existing docs: * **Product and feature briefs.** What a feature does and who it’s for. * **Terminology and glossaries.** Preferred terms, product names, and definitions. * **Architecture overviews.** How your systems fit together. * **Internal FAQs and support notes.** Answers your team already gives customers. Keep each document focused and current. The agent treats every uploaded document as accurate, so remove anything out of date. ## Manage your documents [Section titled “Manage your documents”](#manage-your-documents) The **Uploaded Documents** list shows every document available to the agent. From the list, you can: * **Preview** a document to confirm you uploaded the right file. * **Delete** a document you no longer want the agent to use. Caution Deleting a document is permanent. After you delete it, the agent stops using it in new sessions. ## Verify the agent uses a document [Section titled “Verify the agent uses a document”](#verify-the-agent-uses-a-document) 1. Open the editor and start a new session. 2. Ask the agent a question that only your uploaded document answers. For example, `"What does our onboarding service do?"` 3. Confirm the answer reflects the content of your document. If it doesn’t, check that the document finished processing and appears in the **Uploaded Documents** list. Then start a new session. ## Add a repository note [Section titled “Add a repository note”](#add-a-repository-note) A repository note says what one repository is for, such as what it holds and who reads it. Both agents get every note in every new session. They read a note as a description, not an instruction. 1. Go to **Settings** > **Integrations** > **GitHub Integration**. For a GitLab repository, go to **Settings** > **Integrations** > **GitLab Integration**. 2. Find the repository in your list of saved repositories. A repository without a note shows **Describe what this repository is**. 3. Click the pencil icon on the repository card. 4. Enter a note of up to 5,000 characters, then click **Save note**. A repository you have selected but not yet saved shows **A note can be added once this repository is saved**. Notes apply to sessions started after you save them. Repository notes describe one repository. For conventions that hold across all your repositories, such as spelling or heading style, use [custom instructions](/agent/custom-instructions/). ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### The Knowledge Base tab isn’t available [Section titled “The Knowledge Base tab isn’t available”](#the-knowledge-base-tab-isnt-available) If you don’t see **Settings** > **Organization** > **Knowledge Base**, contact **** with your organization name. ### The upload fails [Section titled “The upload fails”](#the-upload-fails) * Check the file type and size. The upload area shows **That file type isn’t supported.** or **That file is over 3 MB. Upload a smaller file.** when it can’t take a file. * Upload one file at a time. * If the error persists, contact **** with the filename and the error message shown. ### A document shows “Couldn’t process” or “Still processing” [Section titled “A document shows “Couldn’t process” or “Still processing””](#a-document-shows-couldnt-process-or-still-processing) The **Uploaded Documents** list shows the reason next to the document. **Couldn’t process: this file type isn’t enabled for your organization yet** means the file is stored but EkLine can’t read it yet. Contact **** to turn on that file type, then click **Retry**. For any other reason, or a document that stays at **Still processing**, click **Retry**. ### The agent doesn’t use an uploaded document [Section titled “The agent doesn’t use an uploaded document”](#the-agent-doesnt-use-an-uploaded-document) * Confirm the document appears in the **Uploaded Documents** list. * Start a new session. The agent picks up new documents in sessions started after the upload finishes. * If a processed document still isn’t used, reference documents may be off for your organization. Contact ****. ### A repository note isn’t reflected in a draft [Section titled “A repository note isn’t reflected in a draft”](#a-repository-note-isnt-reflected-in-a-draft) * Confirm you saved the note on the correct repository card. * Repository notes apply to sessions started after you save. Start a new session to pick up a change. * Make the note specific about what the repository contains and who reads it. ## Next steps [Section titled “Next steps”](#next-steps) * [Custom instructions](/agent/custom-instructions/): Set the conventions each agent follows in every session. * [Company Profile and Domains](/agent/company-profile/): Describe your company, brand names, audiences and voice. * [Create documentation](/agent/create/): Generate READMEs, API references, and guides. * [Docs Agent reference](/agent/reference/): Supported file types, limits, and FAQ. # Keep documentation screenshots up to date > Set up Docs Agent to recapture stale documentation screenshots from your live product, on demand and on a recurring schedule. The Docs Agent can recapture stale documentation screenshots from your live product. First, run a one-off refresh yourself to confirm the agent captures your product correctly. Then, use an automation to keep the images current. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * Automations and the **Sandbox** tab turned on for your organization. If you don’t see **Automations** in the sidebar, or **Sandbox** under **Settings** > **Docs Agent**, contact . * A repository [connected to EkLine](/agent/github-app-setup/) that holds both your documentation pages and their image files. * A [sandbox](/agent/sandbox/) for each product URL your screenshots come from. The sandbox gives the agent the address to visit and, for pages behind a login, the sign-in variables it needs. * Documentation pages that reference their screenshots. The agent finds candidate images by reading how your pages reference them. ## Step 1: Confirm the sandbox reaches your product [Section titled “Step 1: Confirm the sandbox reaches your product”](#step-1-confirm-the-sandbox-reaches-your-product) The agent can only recapture a screen it can open. Verify this before you schedule anything. 1. Go to your [EkLine dashboard](https://ekline.io/dashboard) and open **Settings** > **Docs Agent** > **Sandbox**. 2. Confirm a sandbox appears under **Configured sandboxes** with the URL your screenshots come from. Note its name: you need the name, not the URL, in Step 3. 3. If no sandbox matches, [create one](/agent/sandbox/#create-a-sandbox). ## Step 2: Run a refresh on demand [Section titled “Step 2: Run a refresh on demand”](#step-2-run-a-refresh-on-demand) Run one refresh yourself before you put it on a schedule. You see what the agent captures, and you find any sign-in or navigation problem while someone is watching. 1. Open the Docs Agent editor and start a new session. 2. Ask the agent to refresh your screenshots against the sandbox by name: ```text Check the screenshots in our documentation against the staging-admin sandbox. Recapture any that no longer match the product, then open a pull request. ``` 3. Watch the [Live browser panel](/agent/editor/#watch-the-agent-browse) as the agent signs in and navigates. Each page it opens appears in real time. 4. Review what the agent reports. It lists the images it recaptured, the screen each one depicts, and any image it deferred with the reason. 5. Click **Raise PR** to open a pull request with the new images. The button prefills a chat message. Send it, and the agent opens the PR. ## Step 3: Schedule a recurring refresh [Section titled “Step 3: Schedule a recurring refresh”](#step-3-schedule-a-recurring-refresh) With the sandbox proven, hand the recurring work to an automation. 1. Click **Automations** in the sidebar. 2. Find **Doc Screenshot Refresh** under the **Accuracy** tab, or enter its name in **Search**, then click **Set up** on its card. 3. In **Sandbox name(s) to screenshot against (one per line)**, enter your sandbox name, for example `staging-admin`. The field starts empty, and the automation cannot turn on until you fill it in. 4. Under **Schedule**, the default is every week on Monday at 9:00 AM. Monthly suits most teams, because screenshots drift more slowly than prose and each run changes at most five images: set **Repeat every** to 1 month and select a day under **Repeat on**. Keep weekly if your interface is under active redesign. 5. Under **Notifications**, leave **Raise a pull request** checked, so the agent delivers each run’s images in a pull request for review. 6. **Send a Slack notification** is also checked. Enter a channel name, such as `docs-team`, and the agent posts a summary of each run there. Or clear the checkbox. 7. Click **Turn on**. ## What the agent changes and what it leaves alone [Section titled “What the agent changes and what it leaves alone”](#what-the-agent-changes-and-what-it-leaves-alone) | The agent | Detail | | -------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Recaptures product screenshots | Images that show your product’s interface, cropped to the element the image depicts | | Creates missing images | When a page’s prose and alt text call for a screenshot that isn’t there | | Replaces only structural changes | A renamed label, a moved control, a new field or a reworked layout: a change a reader would act on | | Skips dynamic content | Row values, timestamps, chart figures, and the signed-in username differ on every visit, so the agent leaves those images alone rather than churning your docs each run | | Skips non-screenshots | Diagrams, logos, illustrations, and mockups stay untouched | | Caps each run | At most five images per run, missing images first. It reports anything it defers | | Changes images, not prose | For documentation that describes the interface incorrectly in text, use the **Doc UI Accuracy Review** template instead | ## Verify the setup [Section titled “Verify the setup”](#verify-the-setup) Confirm both stages are in place: * [ ] A sandbox for each product URL appears under **Configured sandboxes** in **Settings** > **Docs Agent** > **Sandbox**. * [ ] The **Doc Screenshot Refresh** card on the **Automations** page has its toggle on. To see the output before the first scheduled run, click **Edit** on its card, open the **Runs** tab and click **Run now**. Then, under **Recent runs**, click the run’s start time to open the full session. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | --------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The run reports that it couldn’t sign in | The sandbox variables are wrong, or the sign-in form needs a value the sandbox doesn’t have. See [The agent can’t sign in](/agent/sandbox/#the-agent-cant-sign-in). | | The agent browsed the site but stayed signed out | The page host doesn’t match the sandbox URL. A sandbox for `staging.example.com` doesn’t apply to `app.example.com`. | | The run finished without a pull request | Nothing needed replacing. When every candidate image differs only in dynamic content, the agent discards the captures rather than opening an empty pull request. | | A screenshot you expected wasn’t updated | Check the run’s deferred list. Common reasons: the agent couldn’t tell which screen the image shows, the image isn’t a product screenshot, or the run hit its five-image cap. | | The run defers images because a sandbox is not configured | The sandbox name you gave doesn’t match a configured sandbox. Names are fixed at creation, so confirm the exact spelling under **Configured sandboxes**. | | Recaptured images show test data you don’t want published | Point the sandbox at an account with presentable data, then re-run. Review each image in the pull request before merging. | ## Next steps [Section titled “Next steps”](#next-steps) * [Browse authenticated pages with a sandbox](/agent/sandbox/): configure the sign-in variables the agent uses to reach pages behind a login. * [Automations](/agent/scheduled-agents/): explore the full template library for recurring documentation tasks. * [Capture screenshots from a video](/agent/screenshots-from-video/): turn a moment in a recording into a documentation image. * [Crop, redact, and annotate documentation images](/agent/edit-doc-images): make a raw capture publishable before it ships. * [Prevent documentation drift](/agent/prevent-documentation-drift/): catch undocumented code changes alongside stale images. # Generate release notes from completed tickets > Use Docs Agent to turn completed Linear or Jira tickets into structured, user-facing release notes and open a pull request for review. Turn a set of completed tickets into a clear release note for your users. Docs Agent reads each ticket’s title, description, and comments, then rewrites the work in plain language, without copying internal shorthand from the tickets. It groups the work into sections such as Features and Bug Fixes, then opens a pull request you can review before publishing. To produce release notes on a schedule instead, see [Automate release notes](#automate-release-notes). ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * At least one repository [connected to EkLine](/agent/github-app-setup/). * A connected ticket source. Docs Agent generates release notes from **Linear** or **Jira**. Connect one under **Settings** > **Integrations** (see [Integrations](/agent/integrations/)). * The ticket identifiers you want to include, such as `ENG-100` through `ENG-110`, or a milestone or release name the agent can look up. ## Generate a release note [Section titled “Generate a release note”](#generate-a-release-note) 1. Sign in to the [EkLine app](https://ekline.io/dashboard) and click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. 2. In the chat panel, describe the release. Name the tickets, the output sections, and the audience: ```text Generate release notes for v2.1 from Linear tickets ENG-100 through ENG-110. Group them under Features, Improvements, and Bug Fixes. Write for end users, one sentence per item. ``` 3. Wait while the agent reads each ticket and drafts the release note in the editor panel. 4. Review the draft. Ask the agent to refine anything that reads inaccurately, for example `Move the rate-limiting change under Improvements` or `Rewrite the SSO item without internal jargon`. ![The Docs Agent chat panel with a release-notes prompt that names a ticket range and the output sections](/assets/images/docs-agent-release-notes.png) ## Structure the output [Section titled “Structure the output”](#structure-the-output) Tell the agent how to group and order entries so the release note matches your existing format. * Group by type ```text Generate release notes for the tickets in the "1.4" milestone. Group them under Features, Improvements, and Bug Fixes. Within each group, list the most impactful change first. ``` * Match an existing file ```text Generate release notes for tickets ENG-200 through ENG-215. Match the heading style and tone of CHANGELOG.md, and add the new version at the top. ``` * Highlight breaking changes ```text Generate release notes for the v3.0 release from Jira tickets in the PLAT project fixVersion 3.0. Call out breaking changes in a separate section at the top with migration notes. ``` ## Verify the release note [Section titled “Verify the release note”](#verify-the-release-note) Before you publish, confirm the draft is accurate and complete. 1. Enable **View All Changes** in the toolbar to see a diff of what the agent wrote. 2. Check that every ticket you named appears, and that no internal-only tickets slipped in. 3. Confirm version numbers, dates, and section headings match your convention. ## Raise a pull request [Section titled “Raise a pull request”](#raise-a-pull-request) When the draft is ready: 1. Click **Raise PR** in the toolbar. The chat box fills with `Open a pull request for` and your repository’s name. 2. Edit the prompt if you want to add instructions, then press **Enter**. 3. The agent opens the pull request and replies in the chat with a link so you can finish the review on GitHub. To add a late-arriving ticket after the PR is open, ask the agent for the change, then click **Update PR** to push it to the same branch. ## Automate release notes [Section titled “Automate release notes”](#automate-release-notes) To generate release notes on a schedule instead of on demand, turn on the **Release Notes Generator** template in [Automations](/agent/scheduled-agents/). It drafts structured release notes from tickets and GitHub activity since the release notes were last updated, and delivers them as a pull request, a Slack message, or both. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause | Fix | | ------------------------------------------ | ------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- | | The agent can’t find a ticket | The ticket source isn’t connected, or the identifier is wrong | Connect Linear or Jira under **Settings** > **Integrations**, then confirm the ticket key and project. | | Entries read like internal commit messages | The prompt didn’t set an audience | Add `"Write for end users"` and ask the agent to rewrite items in plain language. | | A ticket is missing from the draft | It fell outside the range or milestone you named | Name the ticket explicitly, or widen the range: `"Also include ENG-118"`. | | Sections are in the wrong order | The default grouping doesn’t match your format | Specify the sections and their order in the prompt, or point the agent at an existing changelog file to match. | ## Next steps [Section titled “Next steps”](#next-steps) * [Create documentation](/agent/create/): generate READMEs, API references, and guides from your codebase. * [Update and review](/agent/update-review/): keep documentation in sync with code changes. * [Automations](/agent/scheduled-agents/): automate release notes and other recurring documentation tasks. * [Integrations](/agent/integrations/): connect Linear, Jira, Slack, and more. # Rename a feature across all your docs > Use Docs Agent to rename a feature, change a product term, or update pricing consistently across every page, then ship a single pull request. Make one consistent change across your entire documentation set in a single Docs Agent session. The agent searches the whole repository, updates every occurrence and its variations, shows you the full diff, and opens one pull request for review. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account in your organization. * A documentation repository [connected to EkLine](/agent/github-app-setup/). GitLab works too: [connect GitLab](/agent/gitlab-setup/) instead. * Permission to open pull requests on that repository. ## Step 1: Describe the change and its scope [Section titled “Step 1: Describe the change and its scope”](#step-1-describe-the-change-and-its-scope) Open the editor and tell the agent three things: the old term, the new term, and where the change applies. Name what to leave alone in the same prompt so the agent doesn’t over-reach. 1. Sign in to the [EkLine app](https://ekline.io/dashboard) and click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. 2. In the chat panel, describe the rename and its boundary: ```text We renamed the "Workspace" feature to "Project" across the product. Update every mention in the docs/ folder to use "Project": headings, body text, captions, and prose in code examples. Leave URL slugs, file names, and code identifiers like `workspaceId` unchanged. ``` **Verify:** The agent responds in the chat and begins reading your repository. ## Step 2: Let the agent find every occurrence [Section titled “Step 2: Let the agent find every occurrence”](#step-2-let-the-agent-find-every-occurrence) The agent searches your whole connected repository, not only the files you name. It finds occurrences you might miss, such as a term buried in a troubleshooting page, an image caption, or a comment inside an example. **Verify:** The agent reports the files it plans to change and drafts the edits in the editor panel on the left. ## Step 3: Cover the variations [Section titled “Step 3: Cover the variations”](#step-3-cover-the-variations) A term can appear in multiple forms. Plurals, adjectives, and link text all need to move together, while lookalike words that mean something else must stay. Send a follow-up to catch the variations: * Plurals and forms ```text Also update the variations: "Workspaces" to "Projects" and "workspace-level" to "project-level". Keep the sentence grammar correct after each change. ``` * Link text ```text Update the anchor text of any link that reads "Workspace" or "Workspaces" to match the new term. Leave the link URLs unchanged. ``` * Avoid false matches ```text Don't change the word "workspace" where it refers to a VS Code workspace, which is a different concept. Only rename our product feature. ``` The agent updates the same draft rather than starting over, so each follow-up refines the change in place. ## Step 4: Review the full diff [Section titled “Step 4: Review the full diff”](#step-4-review-the-full-diff) A site-wide change touches many files, so review it as a whole before you ship. 1. Enable **View All Changes** in the toolbar to see a diff of every modification across all files. 2. Scan for the three failure modes of a bulk rename: * **Over-eager matches**: a word that matched the pattern but meant something else. * **Broken links**: anchor text renamed but a target left stale, or the reverse. * **Preserved-term drift**: a code identifier, URL, or filename changed when you asked to keep it. **Verify:** The diff shows the new term throughout your prose, with the identifiers, slugs, and file names you named still intact. ## Step 5: Fix misses with a follow-up [Section titled “Step 5: Fix misses with a follow-up”](#step-5-fix-misses-with-a-follow-up) Reviewing a large change almost always surfaces a straggler. Point the agent at it instead of editing by hand: ```text You missed the term in docs/billing.md, and you changed `workspaceId` in the API example. Revert that identifier. Everything else looks right. ``` The agent applies the correction to the current draft. Repeat until the diff is clean. ## Step 6: Open a single pull request [Section titled “Step 6: Open a single pull request”](#step-6-open-a-single-pull-request) Ship the whole rename as one reviewable change. 1. Click **Raise PR** in the toolbar. 2. The chat box fills with `Open a pull request for` and your repository’s name. Edit it to add a title or description, for example `Rename Workspace to Project across the docs`, then press **Enter**. 3. The agent creates the pull request and responds in the chat with a link to it. **Verify:** The chat shows a link to one pull request that contains every renamed page and nothing unrelated. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause and fix | | ------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | The agent changed a code identifier or URL you wanted to keep | The boundary wasn’t explicit. Ask it to revert that specific change and name what to preserve. For example, tell it to restore the original identifier and leave all code identifiers untouched. | | The agent missed a file | Name the file or section directly, and confirm the path is inside a connected repository. | | Too many changes to review at once | Scope the rename by directory and run it in batches, for example `"Only update docs/api/ for now"`, then repeat for the next directory. | | The old term still appears in a screenshot | A rename changes text, not images. Recapture or edit the image separately with [Keep screenshots up to date](/agent/refresh-doc-screenshots/) or [Edit documentation images](/agent/edit-doc-images/). | ## Next steps [Section titled “Next steps”](#next-steps) * [Customize for your organization](/agent/custom-instructions/): record the new term in your custom instructions so the agent uses it in every future draft. * [Manage a knowledge base](/agent/manage-knowledge-base/): apply the same rename to a connected Confluence or Pylon knowledge base. * [Update and review documentation](/agent/update-review/): keep docs in sync when the code behind a feature changes, not only its name. * [Combine multiple sources into one document](/agent/combine-sources/): pull a ticket, pull request, and spec into a single authoritative page. # Control which pull requests EkLine reviews > Set per-repository review rules to scope which pull requests Docs Agent reviews by branch, title, author, label, or changed files. By default, [automatic PR review](/agent/automatic-pr-review/) assesses every pull request opened into your repository’s default branch, except drafts and, on GitHub, pull requests from a few bot accounts. Review rules let you narrow that scope per repository, so EkLine skips the pull requests you don’t want reviewed. Use review rules to skip draft pull requests, add release branches, or ignore pull requests by title, author, label, or the files they change. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * Automatic PR review on for your organization. It is on by default. * The [EkLine GitHub App connected](/agent/github-app-setup/), or [GitLab connected](/agent/gitlab-setup/), to at least one repository. * **Enable monitoring** turned on for the repository you want to configure. Review rules are only reachable for monitored repositories. ## Open the review rules for a repository [Section titled “Open the review rules for a repository”](#open-the-review-rules-for-a-repository) 1. Log in to your [EkLine dashboard](https://ekline.io/dashboard). 2. Go to **Settings** > **Integrations** > **GitHub Integration** (or **GitLab Integration**). 3. Find the repository in the list and confirm the **Enable monitoring** toggle is on. 4. Click the gear icon next to the repository. Its label reads **Review rules**. The review rules page opens with the repository name at the top and a back link to the integration page. The default rules review every pull request opened into your default branch, except drafts and, on GitHub, pull requests from the bot accounts listed under **Skip PRs from these authors**. ## Choose which pull requests to review [Section titled “Choose which pull requests to review”](#choose-which-pull-requests-to-review) The review rules page groups its controls into filters. Each filter narrows the set of pull requests EkLine reviews: EkLine reviews a pull request only when it passes every filter. ### Skip draft pull requests [Section titled “Skip draft pull requests”](#skip-draft-pull-requests) The **Skip draft PRs** toggle is on by default, so EkLine ignores draft pull requests until you mark them ready for review. Turn it off to review drafts as well. ### Branch rules [Section titled “Branch rules”](#branch-rules) | Field | What it does | | ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- | | **Review PRs into these branches** | EkLine reviews pull requests that target any branch listed here. Your default branch is prepopulated. At least one branch is required. | | **Skip PRs from these branches** | EkLine skips pull requests whose source (head) branch matches an entry. Leave it empty to review pull requests from any branch. | Both fields match branch names with regular expressions. Patterns are **unanchored** unless you add `^` and `$`, and matching is **case-sensitive**. The prepopulated default branch is already anchored, for example `^main$`. To review release branches as well as your default branch, add a pattern such as `release/.*` to **Review PRs into these branches**. To skip documentation-only branches, add `^docs/` to **Skip PRs from these branches**. ### Pull request filters [Section titled “Pull request filters”](#pull-request-filters) | Field | What it does | | -------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Skip PRs whose title matches** | Skips pull requests whose title matches a regular expression, for example `^chore:` to skip chore commits. | | **Skip PRs from these authors** | Skips pull requests opened by the usernames listed. Match usernames exactly. Bot accounts include the `[bot]` suffix, for example `dependabot[bot]`. On GitHub the list starts with `dependabot[bot]`, `renovate[bot]`, `github-actions[bot]` and `ekline[bot]`. | | **Filter by label** | Enter a label name to require it, so EkLine reviews only pull requests carrying that label. Prefix a label with `!` to skip pull requests that carry it, for example `!wip`. | Author matching is case-insensitive. Label matching is case-insensitive on GitHub and case-sensitive on GitLab. ### File filters [Section titled “File filters”](#file-filters) The **Skip PRs that only touch these files** field takes glob patterns. EkLine skips a pull request only when *every* changed file matches at least one pattern. A pull request that touches any non-matching file is still reviewed. For example, `**/*.md` skips pull requests that change only Markdown files, but still reviews a pull request that changes a Markdown file alongside a source file. ## Choose what happens after a review [Section titled “Choose what happens after a review”](#choose-what-happens-after-a-review) The **Output behaviour** section at the top of the page has an **Auto-create documentation PRs** toggle. When it is on, the agent raises the docs PR itself. When it is off, drafts appear in the editor for you to raise yourself. If you don’t see **Output behaviour**, the agent always raises the docs PR. To get the toggle, contact . ## Save your rules [Section titled “Save your rules”](#save-your-rules) 1. Add an entry to any field by typing a value and pressing **Enter** or a comma. Each entry appears as a chip. To remove the last chip, press **Backspace** in an empty field. 2. Click **Save changes**. EkLine validates your patterns and confirms with a **Review rules saved** message. Your changes apply from the next time a pull request is opened or marked ready for review, or, on GitHub, has a label added or removed. A pull request EkLine has already reviewed is not reviewed again. ## Verify your rules [Section titled “Verify your rules”](#verify-your-rules) Confirm the rules take effect on a real pull request: 1. Open a pull request that one of your rules should skip, for example a draft pull request, or one whose title matches a skip pattern. 2. Wait about 30 seconds, then check the pull request. EkLine posts no automatic review comment, because the pull request was filtered out. 3. Open a pull request that your rules should review, for example one into a reviewed branch with a documentation-impacting change. EkLine assesses it and comments when it detects documentation impact. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | ---------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | No gear icon next to the repository | Review rules open only for monitored repositories. Turn on **Enable monitoring** for the repository, then reopen the row. | | Save is blocked with an error on a field | A pattern is invalid or an entry is uncommitted. Fix the highlighted regular expression, or press **Enter** to add a pending entry (or clear it), then save again. | | Save is blocked asking for a branch | **Review PRs into these branches** requires at least one entry. Add your default branch, for example `^main$`, before saving. | | A pull request I expected to skip was still reviewed | Regular expressions are unanchored and case-sensitive. Confirm the pattern matches the exact branch, title, or author. For file filters, remember EkLine skips a pull request only when every changed file matches. | | A pull request I expected to review was skipped | Check each filter in turn: a skip pattern for the branch, title, author, or label may be catching it. | ## Next steps [Section titled “Next steps”](#next-steps) * [Automatic PR review](/agent/automatic-pr-review/): understand the confidence tiers that decide what EkLine does after a pull request passes your rules. * [Prevent documentation drift](/agent/prevent-documentation-drift/): pair automatic PR review with a scheduled drift audit. * [Connect GitHub](/agent/github-app-setup/): install the GitHub App and select repositories to monitor. # Browse authenticated pages with a sandbox > Set up a sandbox so EkLine Docs Agent can sign in to your product and capture screenshots of pages behind a login. A sandbox gives Docs Agent a site URL and the sign-in variables it needs to reach pages behind a login. With a sandbox configured, the agent can open your product, sign in, and capture screenshots of authenticated pages to include in your documentation. Without one, the agent can browse only pages that don’t require signing in. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) * An EkLine organization account. * The **Sandbox** tab under **Settings** > **Docs Agent**. If you don’t see it, see [The Sandbox tab isn’t available](#the-sandbox-tab-isnt-available). * The URL of the site you want the agent to browse, and a set of sign-in credentials for it. ## Create a sandbox [Section titled “Create a sandbox”](#create-a-sandbox) 1. **Open your Docs Agent settings.** Go to your [EkLine dashboard](https://ekline.io/dashboard) and open **Settings** > **Docs Agent** > **Sandbox**. 2. **Name the sandbox.** In **Sandbox name**, enter a unique name such as `admin-staging`. The name identifies the sandbox and can’t be changed later. To rename it, delete and recreate it. Two sandboxes can share a URL with different variables. 3. **Enter the site URL.** In **Sandbox URL**, enter the full URL, such as `https://staging.example.com`. The agent applies this sandbox’s variables when it browses a page on this host. 4. **Add sign-in variables.** Each variable is a name and value pair. EkLine starts you with `email` and `password` rows: enter the values the site needs to sign in. Values for variables named with `password`, `secret`, `token`, or `key` are masked as you type. Click **Add variable** to add more, such as an organization slug. Add at least one variable with a value. 5. **Save the sandbox.** Click **Save sandbox**. The sandbox appears under **Configured sandboxes** with its URL and a badge for each variable name. Caution EkLine stores variable values encrypted, and the **Configured sandboxes** list shows only variable names. Any member of your organization can open a sandbox’s edit dialog, which loads its saved values, so store test credentials only. ## Edit a sandbox [Section titled “Edit a sandbox”](#edit-a-sandbox) 1. **Open the sandbox.** In **Configured sandboxes**, click the edit icon next to the sandbox. A dialog opens with the current URL and variables filled in, titled **Edit** and the sandbox name, such as **Edit admin-staging**. The name is fixed. 2. **Change the URL or variables.** Update the URL, edit values, or add and remove variables as needed. 3. **Update the sandbox.** Click **Update sandbox**. To keep a saved value without retyping it, such as a password, leave that variable’s value blank. EkLine keeps the value already stored. Enter a new value only when you want to replace it. ## Delete a sandbox [Section titled “Delete a sandbox”](#delete-a-sandbox) In **Configured sandboxes**, click the delete icon next to the sandbox. Caution Deleting a sandbox removes its stored variables and can’t be undone. The agent can no longer sign in to that site until you recreate the sandbox. ## Use a sandbox in a session [Section titled “Use a sandbox in a session”](#use-a-sandbox-in-a-session) Ask the agent to browse or screenshot a page on the sandbox’s host. The agent signs in with the sandbox variables, then captures the page. Watch it work in the [Live browser panel](/agent/editor/#watch-the-agent-browse). ```text Sign in to https://staging.example.com and add a screenshot of the billing settings page to the "Manage billing" guide. ``` ## Verify the sandbox [Section titled “Verify the sandbox”](#verify-the-sandbox) * The sandbox appears under **Configured sandboxes** with the correct URL and a badge for each variable. * In a session, ask the agent to screenshot a page that requires signing in. In the **Live browser** panel, confirm the agent reaches the page as a signed-in user. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### The Sandbox tab isn’t available [Section titled “The Sandbox tab isn’t available”](#the-sandbox-tab-isnt-available) If **Settings** > **Docs Agent** has no **Sandbox** tab, contact **** with your organization name. ### The agent can’t sign in [Section titled “The agent can’t sign in”](#the-agent-cant-sign-in) The credentials are wrong, or the sign-in form needs a value you didn’t add. Edit the sandbox, confirm each value is correct, and add any extra variable the login requires, such as an organization slug. ### The agent browses but doesn’t sign in [Section titled “The agent browses but doesn’t sign in”](#the-agent-browses-but-doesnt-sign-in) The page host doesn’t match the sandbox URL. Confirm the host in **Sandbox URL** matches the page you asked the agent to open. A sandbox for `staging.example.com` doesn’t apply to `app.example.com`. ### A saved value seems wrong after editing [Section titled “A saved value seems wrong after editing”](#a-saved-value-seems-wrong-after-editing) You left the value blank while editing, which keeps the previous value. Enter the new value explicitly to replace it. ## Next steps [Section titled “Next steps”](#next-steps) * [Watch the agent browse](/agent/editor/#watch-the-agent-browse): see each page the agent visits in real time. * [Create documentation](/agent/create/#from-screenshots-and-images): Insert screenshots the agent captures into your docs. * [Keep documentation screenshots up to date](/agent/refresh-doc-screenshots/): Recapture stale images from your live product on a schedule. * [Customize Docs Agent for your organization](/agent/custom-instructions/): Set conventions the agent applies to every session. *** ## Stuck? [Section titled “Stuck?”](#stuck) Reply to your welcome email or contact . We read every message. # Automations > Run Docs Agent or Marketing Agent tasks on a schedule: turn on a built-in automation or write your own, choose the agent it runs as, set when it runs, and read what each run did. Automations run Docs Agent or Marketing Agent tasks on a schedule you set, such as a weekly docs drift review, a monthly tutorial gap review or a weekly release marketing brief. [Automations demo](https://www.youtube.com/embed/56ygnpePseA) This video was recorded when automations were called Scheduled Agents, so some screens look different today. ## What an automation is [Section titled “What an automation is”](#what-an-automation-is) An automation is a prompt that EkLine runs on a schedule. Each run starts a new session of the automation’s agent, titled with the automation’s name, and the agent works through the prompt with no one there to answer questions. * **Built-in automations** come from EkLine’s templates. Each has its own card on the **Automations** page. * **Custom automations** run a prompt you write. They are listed under the **Custom** tab. Every automation runs as one agent, the Docs Agent or the Marketing Agent. Each built-in automation has its agent: **Release Marketing Brief** runs as the Marketing Agent, and every other built-in runs as the Docs Agent. For a custom automation, you choose the agent when you create it. The agent cannot be changed afterwards; to run the same prompt as the other agent, create a new automation. If you don’t see **Marketing Agent** under **Runs as**, or the **Marketing** tab, contact . A run uses credits like any other session of its agent, and counts as that agent on the Usage page. When your organization is out of credits, the run is skipped. See [Credits and usage](/workspace/credits-and-usage/). ## What you can automate [Section titled “What you can automate”](#what-you-can-automate) The **Automations** page groups the built-in automations into six tabs: **Accuracy**, **SEO and GEO**, **Content gaps**, **Knowledge base**, **Releases and reports**, and **Marketing**. [Automation templates](/reference/automation-templates/) lists every one, with what it does, its default schedule, and the inputs it asks for. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * For a Docs Agent automation, at least one repository is selected under **Where EkLine writes**, so the agent has docs to work on and somewhere to open pull requests. See [Connect GitHub](/agent/github-app-setup/) or [Connect GitLab](/agent/gitlab-setup/). * To get results in Slack, Slack is connected in **Settings** > **Integrations**. See [Slack](/agent/slack-bot/). Any member of your organization can create, change and delete automations. ## Turn on a built-in automation [Section titled “Turn on a built-in automation”](#turn-on-a-built-in-automation) 1. Click **Automations** in the sidebar. 2. Find the automation. Click a category tab, or enter its name in **Search**. 3. Turn on the toggle on its card to use the defaults, or click **Set up** to review the settings first. With the toggle, the automation turns on with its default schedule and settings, and a message shows when it first runs. If it still needs a value from you, such as your website URL or a Slack channel, its setup panel opens instead. 4. In the setup panel, fill in the automation’s inputs. Use **Additional instructions (optional)** for anything to add to what the automation already does. 5. Set the **Schedule** and **Notifications**, as described below, then click **Turn on**. The first time you turn on an automation that posts to Slack, the panel asks which channel to use. EkLine remembers it and fills it in for your other automations. ## Create a custom automation [Section titled “Create a custom automation”](#create-a-custom-automation) 1. On the **Automations** page, click **New custom automation**. 2. Under **Runs as**, select **Docs Agent** or **Marketing Agent**. The Docs Agent works in your documentation repositories and can open pull requests. The Marketing Agent drafts marketing content from your product, your docs and your company context. 3. Enter a **Name**, up to 200 characters. 4. Write the **Prompt**: what the agent does on every run, up to 10,000 characters. 5. Set the **Schedule** and **Notifications**. 6. Click **Create automation**. Write the prompt so it stands on its own, because no one is there to answer the agent during a run. Name the repository or folder, the audience, the output you want, and how to group or sort it. To run a [skill](/agent/custom-skills/), start the prompt with its slash command, for example `/ekline-docs-drift-review`. * Release notes ```text Generate release notes for all tickets completed since the last release. Organize by Features, Improvements, and Bug Fixes. Write for end users. ``` * Doc audit ```text Review the docs/ directory for outdated content. Flag any references to deprecated APIs or removed features. Create a summary of what needs updating. ``` * README sync ```text Check if the README reflects the current project structure and dependencies. Update any outdated installation instructions or configuration examples. ``` ## How a run builds on earlier runs [Section titled “How a run builds on earlier runs”](#how-a-run-builds-on-earlier-runs) Before a run chooses what to work on, the agent opens the automation’s earlier runs, most recent first, to see what each was asked, what it answered, and the pull requests it opened. Then: * It skips work an earlier run already did or has open for review, names that run and its pull request, and picks something else. * It checks each of those pull requests on GitHub or GitLab rather than trusting the last state EkLine recorded. Work merged into a branch other than your default one, such as `develop`, still counts as done. * When everything it finds is already covered, it says so and finishes without opening a pull request. Each run ends its reply with a record for the next run: * the scope it looked at * what it produced, with the pull request number and title, or why it produced nothing * what it left out, and why * what the next run should start from and check again A run finds earlier runs by the automation’s name. After you rename an automation, its runs from before the rename are not found. When you write in a run’s session yourself, the agent does what you ask, even if an earlier run covered the same work. It first tells you what that work was and where it landed. ## Set the schedule [Section titled “Set the schedule”](#set-the-schedule) Under **Repeat every**, enter a number and select a unit. For every unit except hours, select the hour under **At**. | Repeat every | Options | | ------------ | -------------------------------------------------------------------------------- | | Hours | Every 1 to 24 hours | | Days | Every 1 to 30 days, at a whole hour | | Weeks | Every 1 to 12 weeks, on one or more days under **Repeat on**, at a whole hour | | Months | Every 1 to 12 months, on a day from 1 to 28 under **Repeat on**, at a whole hour | Times use the time zone of the browser you set the automation up in, and the **Schedule** section names it. **Next run** shows when the automation runs next. An interval such as every 2 weeks counts from the day you set the automation up. The run time stays the same on your clock through daylight saving changes. ## Choose where results go [Section titled “Choose where results go”](#choose-where-results-go) The full result of every run is in its session. Under **Notifications** you can also ask for: | Setting | What happens | | ----------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Raise a pull request** | Docs Agent automations only. The agent opens a pull request with its changes when it is done. A run that finds its work already covered by earlier runs opens none. A Marketing Agent automation saves its drafts in its session instead. | | **Send a Slack notification** | The agent posts a summary to the channel you name. | If a pull request or a post does not appear, open the run’s session to see what happened. If Slack is not connected, the agent skips the post and says so in the session. EkLine sends no message of its own when a run starts, succeeds, fails or is skipped, and sends no email. To report EkLine’s work to your team on a schedule, see [Keep your team informed](/agent/keep-your-team-informed/). ## Manage an automation [Section titled “Manage an automation”](#manage-an-automation) Click **Edit** on an automation’s card to open its panel, which has a **Settings** tab and a **Runs** tab. | To | Do this | | ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- | | Run it now | On **Runs**, click **Run now**. The next scheduled run is then counted from this one. | | Skip the next run | On **Runs**, click **Skip next run**. The run after it happens as scheduled. Click it again before the skipped time passes to skip the following run too. | | Change its settings | On **Settings**, make your changes and click **Save changes**. They apply from the next run. | | Turn it off | Turn off the toggle on its card. It keeps its settings and does not run until you turn it back on. There is no separate pause control. | | Delete it | Custom automations only. On **Settings**, click **Delete**, then confirm. This cannot be undone. Sessions from earlier runs stay in your sessions list. | Built-in automations cannot be deleted. Turn one off with its toggle instead. ## Check it worked [Section titled “Check it worked”](#check-it-worked) Open the automation’s **Runs** tab. **Recent runs** lists the five most recent runs, newest first, each with its start time and a status: | Status | Meaning | | ------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Running** | The run is in progress. | | **Succeeded** | The run finished. | | **Failed** | The run stopped with an error. | | **Skipped** | The run did not start a session. The **Runs** tab says why under the run’s start time: your organization was out of credits, or the Marketing Agent is not enabled for your organization. | Click a run’s start time to open its session and see what the agent did, including any pull request it opened. The card on the **Automations** page also shows the latest run’s status. ## Next [Section titled “Next”](#next) * [Automation templates](/reference/automation-templates/): every built-in automation and the inputs it needs. * [Keep your team informed](/agent/keep-your-team-informed/): send a weekly work report and a release brief to Slack. * [Find a past session](/agent/find-sessions/): find a run’s session later. # Capture screenshots from a video or browsing recording > Use Docs Agent to turn a moment in an uploaded video or a browsing recording into a documentation screenshot, then insert it into your docs. Docs Agent turns a moment in a video into a documentation screenshot. Point it at a second of a video and it extracts that frame as a PNG, saves it with your documentation files, and adds the markdown reference in the page you are writing. The video can be one you upload or the recording the agent makes while it browses. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * A repository [connected to EkLine](/agent/github-app-setup/) where the documentation lives. * A source that contains the moment you want to capture, either an uploaded video or a session in which the agent browsed. * For a browsing recording, the **Sandbox** tab under **Settings** > **Docs Agent**. If you don’t see it, contact . An uploaded video must meet these limits: | Requirement | Value | | ------------ | ------------------------- | | Formats | MP4, WebM, MOV, AVI, MPEG | | Maximum size | 500 MB | ## Step 1: Give the agent a source [Section titled “Step 1: Give the agent a source”](#step-1-give-the-agent-a-source) Choose the source that matches where your moment lives. * Uploaded video 1. Sign in to the [EkLine app](https://ekline.io/dashboard) and click **New session** in the sidebar. If the composer shows **Docs** and **Marketing**, select **Docs**. 2. Click the attachment icon in the chat panel and select your video file. 3. Wait for the upload to finish. When you send your message, the agent transcribes the video and reads what happens on screen, which takes a few minutes for a longer recording. The video appears as an attachment in the chat panel once the upload completes. * Browsing recording Some tasks lead the agent to open a browser, for example when you ask it to document a page in your product. The agent records what it visits and keeps the recording with the session, so you can pull a frame from it afterward. 1. Ask the agent to visit the pages you want to illustrate. For example: ```text Open our pricing page and document the plan tiers. ``` 2. Watch the **Live browser** panel in the bottom-right corner of the editor to see the pages it visits. 3. Wait for the agent to finish browsing. The panel header changes to **Recording**, and the recording is available for the rest of the session. To reach pages behind a sign-in, [set up a sandbox](/agent/sandbox/) first. ## Step 2: Find the moment you want [Section titled “Step 2: Find the moment you want”](#step-2-find-the-moment-you-want) Ask the agent where something appears in the video instead of scrubbing through the recording yourself. ```text Which second of this video shows the billing settings page with the plan selector open? ``` The agent answers with a timestamp. Ask follow-up questions if several moments look similar: ```text Is the success message visible at that point, or does it appear later? ``` ## Step 3: Ask for the frame [Section titled “Step 3: Ask for the frame”](#step-3-ask-for-the-frame) Name the second you want and where the image belongs. Telling the agent what the screen shows helps it write useful alt text. * Insert into a page ```text Extract the frame at 47 seconds and insert it into docs/billing/change-plan.md after the "Open billing settings" step. Write alt text describing the plan selector. ``` * Several frames ```text Extract frames at 12, 47, and 90 seconds and place each one next to the step it illustrates in docs/billing/change-plan.md. ``` * From a browsing recording ```text Extract the frame from the browsing recording where the pricing table is fully visible, and add it to the overview page with a caption. ``` The agent extracts the frame, saves it as a PNG in your repository, and adds the markdown reference in the page you named. ## Step 4: Check the image in the editor [Section titled “Step 4: Check the image in the editor”](#step-4-check-the-image-in-the-editor) Select the image’s tab in the editor panel. The image renders inline, with the filename and a status badge below it. Click the preview to open the image larger. Check that: * The frame shows the screen or action you meant to capture. * The image sits next to the step it illustrates. * The alt text describes what the screen shows. If the frame is wrong, reply in the chat with a corrected timestamp: ```text That frame is too early. The dialog is still opening. Extract the frame at 50 seconds instead. ``` ## Step 5: Publish as a pull request [Section titled “Step 5: Publish as a pull request”](#step-5-publish-as-a-pull-request) 1. Enable **View All Changes** in the toolbar to see a diff of what the agent created. 2. Click **Raise PR**. The chat box fills with `Open a pull request for` and your repository’s name. Press **Enter** to send it, or edit it first to add instructions. The agent opens the pull request and replies in the chat with a link. Extracted images commit alongside your documentation changes, so the page and its screenshots land in the same pull request. ## Verify [Section titled “Verify”](#verify) Confirm the capture worked: * [ ] The image file appears in the editor panel and previews inline. * [ ] The page you named has a markdown reference to the new image. * [ ] The pull request includes both the page and the PNG file. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | --------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The agent says it cannot read the video | Check that the file finished processing and is under 500 MB in a supported format. Re-upload it if processing failed. | | The captured frame is a moment too early | The screen was still changing at that second. Ask for a timestamp one or two seconds later. | | The frame from a browsing recording looks low resolution | A recording captures the browser viewport at the recording’s resolution. For a crisp image, ask the agent to capture the page directly with a [sandbox](/agent/sandbox/) instead. | | The agent cannot find a recording for the session | Recordings exist only for sessions in which the agent browsed. Ask it to visit the page first, then request the frame. | | The image reference points at the wrong place in the page | Name the heading or step you want it after, then ask the agent to move it. | ## Next steps [Section titled “Next steps”](#next-steps) * [Create documentation from a demo video](/agent/create-docs-from-video/): turn a whole walkthrough into a written guide. * [Browse authenticated pages with a sandbox](/agent/sandbox/): let the agent sign in to your product and capture pages directly. * [Create documentation](/agent/create/): generate guides, READMEs, and references from your codebase and other sources. # Docs Agent Slack bot > Create documentation drafts directly from Slack by @mentioning EkLine in a channel or thread. Create documentation without leaving Slack. @mention EkLine in a channel the EkLine app is in, or in a thread there, and the bot generates a draft you can review and edit in EkLine. The bot acts only on @mentions. It ignores thread replies that do not mention it, does not answer direct messages, and has no per-channel settings in EkLine. ## Prerequisites [Section titled “Prerequisites”](#prerequisites) Before you begin, you need: * An EkLine organization account. * Slack workspace administrator access to install the app. ## Connect your Slack workspace [Section titled “Connect your Slack workspace”](#connect-your-slack-workspace) 1. Go to **Settings > Integrations** in your EkLine dashboard. 2. Find **Slack** in the integrations list. 3. Click **Connect**. 4. On the Slack page that opens, allow EkLine to access your Slack workspace. 5. Complete the Slack authorization flow. Slack returns you to EkLine. 6. Verify the integration shows as **Connected**. Once connected, anyone in the Slack workspace can mention `@EkLine` to create documentation drafts. ## Create documentation from Slack [Section titled “Create documentation from Slack”](#create-documentation-from-slack) 1. In a Slack channel the EkLine app is in, or a thread there, type `@EkLine` followed by your request. ```text @EkLine create a troubleshooting guide based on this discussion ``` 2. Look for the :eyes: reaction on your message. If your request is unclear, the bot asks a question in the thread. Answer it with a new @mention. 3. Follow the reply in the thread. Docs Agent replies in one of two ways: * **Draft link:** “Got it! I’m looking into this now.”, then ”📝 Draft ready! Review in EkLine” with a link to the session. * **Threaded replies:** a message with a **View EkLine chat** button that opens the session. A typing status shows while Docs Agent works. When it finishes, the agent posts its reply in the thread, with a collapsed **Proposed changes** diff of up to five changed files. Expand it to read the diff in Slack. Every reply ends with a **See the full chat** link to the whole session. To stop a running request, mention `@EkLine stop` in the thread. If you get draft links and want threaded replies, contact . 4. Open the session in EkLine to review and edit the draft. 5. Look for the pull request. With automatic pull requests on, Docs Agent opens one when it changes documentation files. See [Automatic pull requests](#automatic-pull-requests). ## Automatic pull requests [Section titled “Automatic pull requests”](#automatic-pull-requests) When Docs Agent changes documentation files for a Slack request, it opens a pull request with the drafted changes. This is **on by default**. With threaded replies, Docs Agent includes the PR link in its Slack reply. The pull request also appears in the editor session for that draft. If no pull request appears, open the session and click **Raise PR**. ### Disable automatic PRs [Section titled “Disable automatic PRs”](#disable-automatic-prs) If you prefer to review drafts in the editor before creating a PR manually: 1. Go to **Settings** > **Docs Agent** > **Slack triggers** in your EkLine dashboard. 2. Turn off the **Automatically raise PRs from Slack requests** toggle. The setting applies from the next Slack request, including a new @mention in a thread whose session already exists. Turn the toggle back on to resume automatic pull requests. ## Context the bot captures [Section titled “Context the bot captures”](#context-the-bot-captures) | Context type | What the bot captures | | ---------------- | ------------------------------------------------------------------------------------------------------------- | | Thread messages | All messages in the current thread | | Channel messages | Up to 10 messages before your @mention. In a thread, up to 10 before and 10 after the message that started it | | File attachments | Images (JPEG, PNG, GIF, WebP), PDFs, and videos attached to the message | | Reactions | Emoji reactions on messages | ## Write effective prompts [Section titled “Write effective prompts”](#write-effective-prompts) The quality of your prompt determines the quality of the output. | Instead of… | Try… | | ---------------- | ------------------------------------------------------------------------- | | “help with docs” | “Create a troubleshooting guide for the API timeout issue discussed here” | | “document this” | “Write a how-to guide for the workaround Sarah described” | | “make docs” | “Update the authentication docs based on the new flow we agreed on” | **Include specific details:** * Reference the type of documentation you need: tutorial, how-to guide, reference, troubleshooting. * Name specific topics or features from the conversation. * Mention existing documentation to update if applicable. * Point to URLs or ticket IDs for more context. ## Check it worked [Section titled “Check it worked”](#check-it-worked) * A :eyes: reaction appears on your @mention, followed by a reply in the thread that links to EkLine. * The session appears under **Sessions** in the EkLine app. Filter by **Source** > **Slack** to find it. See [Find a past session](/agent/find-sessions/). ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Issue | Solution | | ---------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- | | Bot does not respond | Verify the workspace connection in **Settings > Integrations**. Check that the bot has access to the channel. | | “This workspace isn’t connected to EkLine. Ask an admin to reconnect.” | Ask a workspace administrator to reconnect the integration in EkLine settings. | | Draft does not match expectations | Give more specific instructions in your mention. Include the documentation type and key topics to cover. | | Bot takes longer than expected | Complex requests may take longer. The bot notifies you if a request times out. | | Bot responds but link is broken | Refresh the page and try the link again. If the issue persists, contact ****. | ## Next steps [Section titled “Next steps”](#next-steps) * [Turn a Slack support thread into a troubleshooting doc](/agent/slack-thread-to-troubleshooting-doc/): Convert a resolved support conversation into a published guide. * [GitHub PR bot](/agent/github-integration/): Trigger Docs Agent from pull request comments and code reviews. * [Create documentation](/agent/create/): Generate READMEs, API references, and guides from your codebase. * [Update and review](/agent/update-review/): Keep documentation in sync with code changes. * [Integrations](/agent/integrations/): Connect Notion, Linear, Jira, and other tools. # Turn a Slack support thread into a troubleshooting doc > Use the Docs Agent Slack bot to convert a resolved support thread into a troubleshooting guide and open a pull request for review. Turn a resolved Slack support thread into a published troubleshooting doc without leaving Slack. @mention EkLine in the thread, and Docs Agent reads the conversation, drafts a troubleshooting guide, and opens a pull request you review and merge. ## Before you begin [Section titled “Before you begin”](#before-you-begin) You need: * An EkLine account with Docs Agent enabled. * The [Docs Agent Slack bot](/agent/slack-bot/) connected to your workspace. * At least one documentation repository [connected to EkLine](/agent/github-app-setup/). * A resolved support thread: one where the problem and the fix are both visible in the conversation. ## Convert the thread [Section titled “Convert the thread”](#convert-the-thread) 1. Open the Slack thread where the team diagnosed and resolved the issue. The clearer the thread states the symptom and the fix, the better the draft. 2. Reply in the thread with an `@EkLine` mention that names the document type, the problem, and where the doc should live: ```text @EkLine turn this thread into a troubleshooting guide for the API timeout error. Add it to the docs repo under troubleshooting, and describe the symptom, the cause, and the fix. ``` The bot reacts with 👀 to confirm it picked up your request, then starts working. 3. Follow the bot’s progress in the thread. Docs Agent reads the thread, identifies the problem and the resolution, and drafts the guide. When it starts, the bot posts a message with a **View EkLine chat** button. Click it to open the session in the editor, where you can refine the draft. Each reply from the agent ends with **See the full chat**, a link to the same session, so you can reopen it from anywhere in the thread. 4. Review the draft. Docs Agent posts the diff in the thread as a file titled **Proposed changes**, which Slack shows collapsed until you expand it, so you can read the changes without leaving Slack. ## Write an effective mention [Section titled “Write an effective mention”](#write-an-effective-mention) A support thread is chatty, so tell the bot exactly what to produce. A specific mention keeps off-topic messages out of the draft. | Instead of… | Try… | | ----------------------- | --------------------------------------------------------------------------------------------- | | “@EkLine document this” | “@EkLine write a troubleshooting guide for the failed webhook delivery issue in this thread” | | “@EkLine make a doc” | “@EkLine turn this thread into a troubleshooting entry: symptom, cause, and resolution steps” | | “@EkLine help” | “@EkLine document the workaround Priya described, and link the related Jira ticket” | **Include specific details:** * State the document type: troubleshooting guide, how-to, or FAQ entry. * Name the error or symptom the thread resolved. * Point to the repository and folder where the guide belongs. * Reference ticket or pull request URLs mentioned in the thread for extra context. ## What the bot reads from the thread [Section titled “What the bot reads from the thread”](#what-the-bot-reads-from-the-thread) Docs Agent gathers context from the conversation to produce an accurate guide: | Context type | What the bot captures | | ---------------- | ----------------------------------------------------------------- | | Thread messages | Every reply in the thread | | Channel messages | Messages surrounding the thread for extra context | | Attachments | Images (JPEG, PNG, GIF, WebP), PDFs, and videos up to 100 MB each | Docs Agent keeps only the attachments relevant to your request and ignores the rest, so it picks up screenshots of the error while leaving unrelated files out. ## Review and merge [Section titled “Review and merge”](#review-and-merge) 1. Read the diff in the thread, or click **View EkLine chat** to open the draft in the editor. 2. Ask Docs Agent to refine anything that reads inaccurately. Reply in the thread with a fresh `@EkLine` mention, for example `@EkLine add a prevention section` or `@EkLine tighten the resolution steps to three bullets`. 3. When the draft changes documentation files, Docs Agent opens a pull request automatically and includes the link in its reply. Open the pull request to finish the review on GitHub. 4. Merge the pull request to publish the troubleshooting guide. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) | Problem | Cause | Fix | | ------------------------------------- | ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | The bot doesn’t respond | The bot acts only on `@EkLine` mentions | Reply in the thread with a fresh `@EkLine` mention. Plain replies without a mention are ignored. | | The bot asks a clarifying question | Your request was ambiguous | Answer with another `@EkLine` mention. The bot asks only once: on your next mention, Docs Agent starts the draft with its best reading of the request. | | The draft mixes in off-topic messages | The mention was too general | Name the specific error and document type in your mention so the bot filters the conversation. | | No pull request appears | The turn didn’t change documentation files, or **Automatically raise PRs from Slack requests** is off | Confirm the draft edits docs files. Check the **Automatically raise PRs from Slack requests** toggle under **Settings** > **Docs Agent** > **Slack triggers**. | | An attachment was ignored | The file is over 100 MB, an unsupported type, or unrelated to your request | Re-share the file under 100 MB in a supported format, and reference it in your mention. | | The guide lands in the wrong place | Docs Agent chose a repository or folder you didn’t intend | Name the target repository and path in your mention, then ask the bot to move the file. | ## Next steps [Section titled “Next steps”](#next-steps) * [Docs Agent Slack bot](/agent/slack-bot/): connect the bot and learn how mentions, drafts, and diffs work. * [Create documentation](/agent/create/): generate docs from your codebase, videos, and other sources. * [Update and review](/agent/update-review/): keep documentation in sync with code changes. * [Integrations](/agent/integrations/): connect Slack, Notion, Jira, and other tools. # Update and review documentation > Keep documentation in sync with code changes and get AI feedback on existing content. Ask the Docs Agent to update documentation that exists but needs changes, or to review it and give feedback without changing it. ## Update from tickets [Section titled “Update from tickets”](#update-from-tickets) When a feature ships or a bug is fixed, update documentation by referencing the ticket. The agent reads the ticket details and identifies which docs need changes. **Works with:** Linear, Jira * Basic update ```text Update the documentation based on Linear ticket ENG-1234. ``` * With context ```text ENG-1234 added rate limiting to the API. Update the API reference with the new rate limit headers and error responses. ``` * Multiple tickets ```text Update the authentication docs based on ENG-1234, ENG-1235, and ENG-1236. ``` ## Update from code changes [Section titled “Update from code changes”](#update-from-code-changes) Point the agent to code changes and it updates relevant documentation. ```text The login flow changed in the last commit. Update the authentication guide to reflect the new OAuth implementation in src/auth/. ``` ```text We deprecated the /api/v1/users endpoint. Update the API reference to mark it as deprecated and point to the v2 replacement. ``` ## Update specific sections [Section titled “Update specific sections”](#update-specific-sections) Target specific parts of a document instead of regenerating everything. ```text Update only the "Configuration" section in docs/getting-started.md with the new environment variables from .env.example. ``` ```text Add a "Troubleshooting" section to the deployment guide covering the three most common issues from our support tickets. ``` ## Review documentation [Section titled “Review documentation”](#review-documentation) Get AI feedback on existing documentation without making changes. The agent analyzes your content and provides suggestions. ### What the agent checks [Section titled “What the agent checks”](#what-the-agent-checks) | Aspect | What it looks for | | ---------------- | ------------------------------------------------------------ | | **Clarity** | Can a reader follow the instructions? Is the language clear? | | **Completeness** | Are steps missing? Are edge cases covered? | | **Accuracy** | Does the documentation match the current code? | | **Consistency** | Does terminology match other docs? Is formatting consistent? | ### Example prompts [Section titled “Example prompts”](#example-prompts) ```text Review the getting started guide and suggest improvements for clarity and completeness. ``` ```text Review the API reference for the authentication endpoints. Check if the examples still work with the current code. ``` ```text Review docs/deployment.md from the perspective of a developer who has never deployed this application before. ``` ### Applying suggestions [Section titled “Applying suggestions”](#applying-suggestions) ![Docs Agent review checkpoint listing completeness, technical accuracy, and style compliance feedback in the chat panel](/assets/images/docs-agent-review-checkpoint.png) The agent provides feedback in the chat panel. You can: 1. Apply suggestions you agree with by asking the agent to make the change 2. Edit directly in the editor panel 3. Iterate by asking for more specific feedback ## Verify before publishing [Section titled “Verify before publishing”](#verify-before-publishing) ### View changes [Section titled “View changes”](#view-changes) Enable **View All Changes** in the toolbar to see a diff of all modifications. Review each change before creating a pull request. Look for: * Unintended deletions. * Formatting issues. * Content that needs adjustment. ### Link checks [Section titled “Link checks”](#link-checks) Before it finishes, the agent checks that the internal links in its drafts point to pages and headings that exist. Check links to external sites yourself before you raise the pull request. ### Create a pull request [Section titled “Create a pull request”](#create-a-pull-request) When satisfied with the changes: 1. Click **Raise PR** in the toolbar. The button prefills a prompt in the chat panel: `Open a pull request`, or, when you raise one for a single repository in a session with several, `Open a pull request for docs-site` with that repository’s name. 2. Edit the prompt if you want to add instructions, then press **Enter** to send it. 3. The agent creates the pull request and responds in the chat with a link to the PR. If your session spans multiple repositories, the toolbar shows a **Linked PRs** menu in place of the **Raise PR** button. Its button reads **PRs (N)**, or how many repositories need action. Open it and click **Raise PR** next to the repository you want. ### Update an existing pull request [Section titled “Update an existing pull request”](#update-an-existing-pull-request) After a PR is open, you can push more changes to it: 1. Make further edits in the editor or ask the agent for more changes in the chat. 2. Click **Update PR** in the toolbar when new changes are ready. The agent prefills a prompt like `Update pull request #42 with my latest changes`. 3. Edit the prompt if needed, then press **Enter**. 4. The agent commits and pushes your changes to the existing PR branch. ## Next steps [Section titled “Next steps”](#next-steps) * [Create documentation](/agent/create): generate READMEs, API references, and guides from your codebase. * [Automations](/agent/scheduled-agents): automate recurring doc tasks such as drift reviews and screenshot refreshes. * [Integrations](/agent/integrations): connect Slack, Notion, Linear, Jira, and other tools. * [Set up automated review](/reviewer/quickstart): run EkLine Docs Reviewer in your CI/CD pipeline to catch issues before they merge. # Visual style guidelines > Set the brand colors and look that EkLine's agents use when they draw hero images, social cards, diagrams and section illustrations. Visual style guidelines tell EkLine’s agents which colors and look to use when they draw an image for your docs or content. ## Before you begin [Section titled “Before you begin”](#before-you-begin) * You set visual style guidelines in Company Profile. Any member of your organization can edit it. See [Company Profile and Domains](/agent/company-profile/). * Your first AI report drafts them from your website. See [What onboarding fills in](/agent/company-profile/#what-onboarding-fills-in). * **Refresh from website**, under the field, drafts them again from the website saved under **Your sites**. See [Refresh Visual Style Guidelines from your website](/agent/company-profile/#refresh-visual-style-guidelines-from-your-website). * The Marketing Agent always uses them. The Docs Agent uses them only when company context is on for your organization. To check or change this, contact ****. ## Where they are used [Section titled “Where they are used”](#where-they-are-used) Both agents draw each image as shapes, text and diagrams, then save it as a PNG file. They never produce photographs. | Agent | When it draws an image | | --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Marketing Agent | For a blog post, a cover and an inline diagram; for a LinkedIn post, one image when it strengthens the post; for a newsletter, section images when a section is clearer with one. At most three images per draft. You can ask for a different image, a change, or none. | | Docs Agent | When a page needs a picture that is not a Mermaid diagram, such as an illustration, or a diagram on a site that does not render Mermaid. | The agents take your colors and style notes from the **Visual Style Guidelines** field and nowhere else. With the field empty, or with no colors in it, images use a neutral palette. ## What to write [Section titled “What to write”](#what-to-write) Write two parts: your colors, then a few sentences about the look. **Colors.** Put each color on its own line, with a role and a hex value such as `#2563eb` (a CSS color name such as `navy` also works). The Docs Agent fills five color roles from these labels: | Label you write | What it colors | | ------------------------------ | -------------------------------------------------------------------------------------- | | `Primary`, `Brand` or `Accent` | The accent: one focal element, one emphasized line, badges and arrowheads | | `Secondary` | The accent when there is no `Primary`, `Brand` or `Accent` color; otherwise a fallback | | `Dark` | The background of a dark image; boxes and strokes on a light one | | `Background` | The background of a light image | | `Light` | Text on a dark image; card fills on a light one | | `Text` or `Ink` | Body text on a light image | If two colors would be hard to read together, the agent swaps one for another color from your list or a neutral one. **Style notes.** In plain sentences, describe the look: which shapes and motifs, how busy or empty, line weight, rounded or square corners, flat fills or a soft glow, and whether to use gradients. A Docs Agent image never follows a note that asks for a named font, a logo or other file, photographs, people, or faces, or a color outside the five roles. It also can’t include logos, uploaded images, or named fonts. A Marketing Agent image draws your logo and the fonts a DESIGN.md guide names, but never uploaded images, photographs, or people. The Marketing Agent tells you in chat which notes an image does not follow, and why. ### Example [Section titled “Example”](#example) Acme, a developer tool company with a blue brand, might write: ```text Primary: #2563eb Dark: #0f172a Background: #f8fafc Text: #1e293b Flat, geometric illustrations with generous whitespace. Rounded corners and thin lines. No gradients. ``` ### A DESIGN.md guide [Section titled “A DESIGN.md guide”](#a-designmd-guide) The field can also hold a design guide in DESIGN.md format, which is what onboarding and **Refresh from website** write from your website. You can edit it or paste your own. Give each color one of the roles in [What to write](#what-to-write). The Marketing Agent reads your colors, fonts, and logo best from this format. From color lines, it takes one brand color, your `Accent`, `Primary`, or `Brand` color, and uses neutral colors for the rest. Its logo comes from the guide’s **Logo** section, or else from your organization’s logo in **Settings**. ## Steps [Section titled “Steps”](#steps) 1. Go to **Settings** > **Organization** > **Company Profile**. 2. Find the **Visual Style Guidelines** field, near the end of the form. 3. Write your colors, one per line with a role, then your style notes, or paste a DESIGN.md guide. To draft a guide from your website instead, select **Refresh from website** under the field, check the preview, and choose **Replace guide**. The field holds up to 10,000 characters. 4. The field saves when you stop typing or move to another field, and **Saved** appears under it. The change applies to images drawn in sessions started after you save. ## Check it worked [Section titled “Check it worked”](#check-it-worked) 1. Start a new Marketing Agent session and ask for a short blog post with a cover image. 2. Open the image in the draft and check that it uses your colors and style. 3. Read the agent’s summary in chat. It lists any style note the image does not follow. ## Troubleshooting [Section titled “Troubleshooting”](#troubleshooting) ### Images use neutral colors instead of yours [Section titled “Images use neutral colors instead of yours”](#images-use-neutral-colors-instead-of-yours) * Put each color on its own line, with a role label and a hex value, as in the [example](#example). * If your guide has no colors, select **Refresh from website** under the field to read them from your site, or add them yourself. * Start a new session after you save. A session that was already open keeps the guidelines it started with. * The Docs Agent uses them only when company context is on for your organization. See [Before you begin](#before-you-begin). ### The Visual Style Guidelines field isn’t there [Section titled “The Visual Style Guidelines field isn’t there”](#the-visual-style-guidelines-field-isnt-there) The field is part of Company Profile. If you don’t see **Settings** > **Organization** > **Company Profile**, contact **** with your organization name. ## Next [Section titled “Next”](#next) * [Company Profile and Domains](/agent/company-profile/): Fill in the rest of your company’s profile. * [Custom instructions](/agent/custom-instructions/): Add standing instructions for one agent. # Instructions, notes, skills, memory and rules: which to use > Compare every setting that gives EkLine's agents context: where it lives, which agent reads it, and when to use it. EkLine has several places to tell its agents about your company and how to work, and a few settings with similar names that do different jobs. A change to a setting the agents read applies to sessions started after you save it. A session that is already open keeps what it started with. ## Every setting at a glance [Section titled “Every setting at a glance”](#every-setting-at-a-glance) | Setting | Where it lives | What it is for | Who reads it | Use it when | | -------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------- | | [Custom instructions](/agent/custom-instructions/) (Docs Agent) | **Settings** > **Docs Agent** > **Custom instructions** | How to write docs: spelling, terms, structure, audience. Up to 20,000 characters. | Docs Agent only | Every docs session should follow a convention. | | [Custom instructions](/agent/custom-instructions/) (Marketing Agent) | **Settings** > **Marketing Agent** > **Custom instructions** | How to write posts: voice, person, how posts open and close. Up to 20,000 characters. | Marketing Agent only | A convention is for posts, not docs. | | [Company Profile](/agent/company-profile/) | **Settings** > **Organization** > **Company Profile** | Facts about your company: product, website, brand names, documentation sites, audiences, personas, competitors, tone and voice. | Marketing Agent always. Docs Agent when company context is on for your organization. | Both agents should know who you are and who you write for. | | Visual Style Guidelines | The **Visual Style Guidelines** field in **Company Profile** | Colors and style notes for the images the agents draw. | Both agents, when they create an image (the Docs Agent only when it reads Company Profile) | Images should match your brand. | | [Reference documents](/agent/reference-documents/) | **Settings** > **Organization** > **Knowledge Base** | Files the agents read for facts, such as briefs and glossaries. Up to 3 MB each. | Both agents, when reference documents are on for your organization | The facts are in a document, not in your code. | | [Repository notes](/agent/reference-documents/) | **Settings** > **Integrations** > **GitHub Integration** or **GitLab Integration**, on each saved repository | What one repository is, read as a description, not an instruction. Up to 5,000 characters. | Both agents | An agent needs to know what a repository holds. | | [Skills](/agent/custom-skills/) | **Settings** > **Docs Agent** > **Skills** | A workflow you run with its slash command, or that the agent uses when a task matches. | Both agents, though the settings page names only the Docs Agent | You repeat the same multi-step task. | | Rules | **Settings** > **Docs Agent** > **Rules** | Standing rules, each **General** (always in force) or **Review** (applied before a docs pull request is opened or updated), optionally for some files only. | Both agents, though the settings page names only the Docs Agent | A requirement must always hold. | | Organization memory | Nowhere in the app | Notes the agents keep for themselves while they work, shared by every session in your organization. No page shows it. | Both agents | You do not set it. For a standing rule, use custom instructions or Rules. | | Reviewer Rules | **Settings** > **Docs Agent** > **Reviewer Rules** | Turns the Docs Reviewer’s built-in style checks on or off. | The [Docs Reviewer](/reviewer/overview/). The agents don’t read it. | You want the Docs Reviewer to check more or less. | | Terminology | **Settings** > **Organization** > **Terminology** | Terms to avoid, each with the term to use instead. | The [Docs Reviewer](/reviewer/overview/). The agents don’t read it. | Reviews should flag a term. | | Dictionary | **Settings** > **Organization** > **Dictionary** | Words the Docs Reviewer’s spell check accepts. | The [Docs Reviewer](/reviewer/configuration/dictionary/). The agents don’t read it. | Reviews flag your product names as misspelled. | | [Automations](/agent/scheduled-agents/) | **Automations** in the sidebar | A prompt and a schedule. Each run starts a Docs Agent session with that agent’s usual context. It is not context itself. | Starts Docs Agent sessions only | A task should run on a schedule. | | [Review rules](/agent/review-rules/) | **Settings** > **Integrations** > **GitHub Integration** or **GitLab Integration**, the **Review rules** icon on a repository | Which pull requests automatic PR review looks at, by branch, author, title, label or file. | Automatic PR review | EkLine should skip some pull requests. | If a tab is missing from your settings, or you want company context or reference documents on, contact . ### Three settings called rules [Section titled “Three settings called rules”](#three-settings-called-rules) * **Rules** tell the agents what to do. A rule of type **Review** applies before a docs pull request is opened or updated. * **Review rules** choose which pull requests [automatic PR review](/agent/automatic-pr-review/) looks at. * **Reviewer Rules** turn the Docs Reviewer’s style checks on and off. The agents don’t read them. ## Examples [Section titled “Examples”](#examples) **Acme wants every docs draft to say “sign in”, never “log in”.** Add the instruction to the Docs Agent’s custom instructions. To have the Docs Reviewer flag “log in” in pull requests too, add it to **Terminology**. **Acme wants both agents to know that its main competitor is Hooli.** Add Hooli under **Competitors** in Company Profile. **Every pull request to Acme’s documentation that touches the API reference must update the changelog.** Add a rule of type **Review**, with `docs/api/**` in **Applies to (optional)**. When the agent can’t meet a rule, it says so in the pull request body. **Acme drafts release notes the same way for every version.** Write the steps once as a skill named `Release notes checklist`, and run it by typing `/release-notes-checklist`. ## FAQ [Section titled “FAQ”](#faq) ### Where do I see the contents of a skill? [Section titled “Where do I see the contents of a skill?”](#where-do-i-see-the-contents-of-a-skill) In **Settings** > **Docs Agent** > **Skills**. The list shows each skill’s name, command and **When to use it** text. Click a skill’s edit icon to see its instructions. The editor has no list of skills. ### Do Rules and Skills apply to the Marketing Agent? [Section titled “Do Rules and Skills apply to the Marketing Agent?”](#do-rules-and-skills-apply-to-the-marketing-agent) Yes. Both agents use them in every new session, even though their settings pages name only the Docs Agent. A rule of type **Review** comes into play when an agent opens or updates a docs pull request. ### Do the agents use my Terminology list and Dictionary? [Section titled “Do the agents use my Terminology list and Dictionary?”](#do-the-agents-use-my-terminology-list-and-dictionary) No. Both configure the Docs Reviewer. To make an agent use or avoid a term, put it in that agent’s custom instructions or in a rule.