For AI agents

If you are an AI coding agent setting Patchrooms up for your user, start here. This section is the short path: provision a project, then run the feedback→fix loop. The rest of the docs are written for humans clicking through the dashboard — you don’t need them unless something below links there.

Human vs Agent

Pages meant for a person doing it by hand live under Start here. Pages meant for you to execute live under For agents (this section) and carry an Agent badge. When in doubt, follow this page’s tree.

The tree

1. Set up the project Take a setup token, configure widget + channels + URL matching over REST, mint a narrow feedback key, leave a patchrooms-feedback skill, revoke the setup token.

2. Run the loop Workflow modes: one-shot review, test pass, a timed loop that auto-fixes, spec-writing, or accumulate-and-ask. Pick by what the user wants.

Read & triage feedback MCP tools list_reports / get_report / set_status, plus a no-MCP curl fallback. The operational key carries feedback:read + feedback:write.

Configure over REST /api/v1 with pr_sk_ bearer keys: project config, channels, key minting. The setup flow above uses these.

Concepts you'll touch Projects, channels (report routing), reports (blocks: text/screenshot/voice/selection), exports.

Write more skills Agents writing agents — the SKILL.md format, and vetted skill sources to build on.

Machine-readable index

• Curated link tree: llms.txt — this site, one file.
• Full sitemap: sitemap-index.xml.
• Marketing site index: patchrooms.com/llms.txt.

What to tell your user

After setup, hand the user a short summary: which channels exist, what URLs are captured, that the widget is embedded, and how they file feedback (click an element in the running app, leave text/screenshot/voice). Then they comment, you pull & fix.