I got tired of spending thousands of dollars a year on SaaS tools just to keep our documentation alive. One subscription to record click-by-click walkthroughs. Another to annotate the screenshots. A knowledge-base platform to host the result. And the screenshots still went stale the moment we shipped a UI change, so we paid the tools and paid the time to redo the work.
Meanwhile, the same coding agents that ship our features can already open a browser, click through a flow, and see exactly what a user sees. So the question got hard to ignore: if the agents write the code, why can't they document it too?
That is Guidewright: an open-source Agent Skill that produces and maintains end-user documentation by driving your live product, not by reading your code. It replaced a pile of subscriptions with one skill any coding agent can run.
The documentation tax
Documentation is one of those costs that hides in plain sight. The tools are recurring. The labor is recurring. And the output decays: a guide is only correct until the next release renames a button or moves a setting. You are not buying documentation once, you are renting it forever.
The insight behind Guidewright is that the decay problem and the cost problem have the same fix. If re-documenting a flow is just re-running an agent, then keeping docs current stops being a budget line and becomes a command. The same approach we use to ship features with an agent works for the docs that describe them.
Two skills that work as a pair
guidewright-captureDrives your app through Chrome DevTools MCP, screenshots each step with a red box on the exact element to click, and authors or refreshes the how-to page.
You get: An annotated, click-by-click walkthrough that matches the live UI
guidewright-reviewWalks the path an existing doc describes in the live product and returns a prioritized review (Blocker, Major, Minor) with paste-ready fixes.
You get: A specific critique that catches stale screenshots, renamed buttons, and dead ends
Capture makes the walkthrough. Review holds it to a standard distilled from established practice: the Diataxis framework, Every Page Is Page One, minimalist instruction, Nielsen Norman Group reading research, and the Google and Microsoft style guides.
Why it drives the live app
Most doc tooling works from a recording or from the code. Guidewright works from the running product, and that changes the quality of the output in two concrete ways.
The red box is real. Instead of painting a rectangle onto a finished PNG, capture renders the highlight into the live page using the target element's own bounding rectangle, the same handle it used to click. The box lands dead-on and scales with the screenshot, so it points at the actual button a user needs, not an approximate spot.
The review actually walks the path. A read-only review can only judge the words on the page. Guidewright's review opens the product and follows the documented steps, so it catches the failures that matter: a screenshot that no longer matches, a button that got renamed, a step that is now missing, a flow that dead-ends. When the UI changes, you re-run the skill and the guide updates itself.
Install
Guidewright follows the open Agent Skills standard, so the same skill is portable across every supported agent. The npx skills CLI auto-detects which agents you have installed and drops the skills into the right place. From any project:
$ npx skills add TurboDocx/guidewright$ npx skills add TurboDocx/guidewright --skill guidewright-capture$ npx skills add TurboDocx/guidewright --skill guidewright-reviewGuidewright needs the Chrome DevTools MCP server, which the skills use to drive the browser. Once installed, just ask: “document how to create a template” runs capture, and “review this guide for clarity” runs review.
Supported agents
Plus any other tool that supports the Agent Skills standard. Both skills are product-agnostic: point them at any app and any docs site. They assume a Docusaurus site by default and adapt to other setups.
Documentation is the next thing agents take off your plate
We already trust agents to write code, open pull requests, and run tests. Documentation is the obvious next step, because it is exactly the kind of work that is high-effort, low-glamour, and constantly out of date. The thing standing in the way was never capability. It was that most tools made the agent guess from a recording or a description.
Give the agent the live product instead, and the guesswork disappears. It clicks the real button, captures the real screen, and writes the step. That is the whole idea, and it is why we shipped Guidewright as open source under the MIT license. Point it at your own product and stop paying a subscription to keep your guides honest.
Frequently asked
What is Guidewright?
Guidewright is an open-source Agent Skills plugin with two skills, guidewright-capture and guidewright-review, that write and maintain product documentation by driving your live app through the Chrome DevTools MCP server. Capture screenshots each step with a red box on the exact element to click; review walks the documented path and returns prioritized fixes.
How is Guidewright different from a screen recorder or screenshot tool?
It drives the running product instead of working from a recording or from your code. The red box is rendered into the live page from the target element’s own bounding rectangle, so it lands exactly on target and scales with the screenshot. Review actually walks the path, which catches stale screenshots, renamed buttons, missing steps, and dead ends a read-only pass never sees.
Which AI coding agents work with Guidewright?
Claude Code, GitHub Copilot in VS Code, Cursor, OpenCode, OpenAI Codex CLI, and Gemini CLI. Any tool that supports the open Agent Skills standard from agentskills.io. It also needs the Chrome DevTools MCP server, which the skills use to drive the browser.
How do I install Guidewright?
Run "npx skills add TurboDocx/guidewright" in any project to install both skills. To add just one, use --skill guidewright-capture or --skill guidewright-review. The CLI auto-detects which AI agents you have installed and drops the skills into the right config directory.
Does Guidewright work with any product and docs site?
Yes. Both skills are product-agnostic, so you can point them at any app and any docs site. They assume a Docusaurus docs site by default and adapt to other setups.
Point it at your product
Install Guidewright, then ask your agent to document a feature or review an existing guide. The walkthrough writes itself.
$ npx skills add TurboDocx/guidewrightRelated reading
Guidewright Launch Announcement →
The official press release in the TurboDocx newsroom.
AI Engineering Guide →
Build features with AI coding agents, end to end.
Install the TurboDocx SDK via Agent Skill →
Our other open-source skill: integrate in one prompt.
How to Build a Claude Code Skill →
Author your own agent skill from scratch.
Best Claude Code Skills & Plugins →
The tools worth installing for AI development.
CLAUDE.md Guide →
Give your agent the context to do great work.
TurboDocx API & SDK →
The full developer surface for document automation.