The Tailr plugin

This directory is the installable bundle: skills, the MCP server, and a manifest for each agent that has a plugin format. It is not part of the npm package — files in package.json keeps it out of the tarball.

The catalogs that point here live at the repository root:

plugin/
  .claude-plugin/plugin.json    Claude Code
  .cursor-plugin/plugin.json    Cursor
  .codex-plugin/plugin.json     Codex
  plugin.json                   Agent Plugins / Copilot
  .mcp.json                     Claude, Codex, Copilot (`tailr mcp`)
  mcp.json                      Cursor / Agent Plugins (same server, stdio typed)
  skills/start/SKILL.md         /tailr:start — `tailr start` / `tailr_start`, hand over the URL
  skills/config/SKILL.md        /tailr:config — read or change the reviewer's settings
  skills/review/SKILL.md        the operating rules for the review loop
  cursor-commands/*.md          the same commands, as Cursor slash commands
  rules/review.mdc              the same loop, as a Cursor rule (not a slash command)

/tailr:start (and Cursor's /start) tell the agent to run npx … tailr start or call tailr_start — Tailr detaches itself. Agents must not shell-background a bare tailr, and must not edit anything under .tailr/.

The repository root also carries Claude and Cursor plugin manifests that point into this directory. That is what the in-app directories read when someone submits gcrft123/tailr — they look for a plugin at the repo root, not a marketplace wrapping one. Marketplace installs still copy only plugin/.

Gemini CLI clones the whole repository as an extension, so skills/ at the repo root is a copy of plugin/skills/ — Gemini only discovers skills next to gemini-extension.json.

A skill marked disable-model-invocation: true is a command: the user invokes it by name and the model never reaches for it on its own. Those are the ones scripts/plugin.js copies into cursor-commands/, where Cursor reads them as /<name> rather than /tailr:<name>. Adding one is adding the skill; --sync generates the Cursor copy and --check fails if it drifts.

What it is for

tailr init makes the rules durable by writing them into the project's AGENTS.md / CLAUDE.md and registering the MCP server in its .mcp.json. That works, but it edits four files in someone's repository to do it.

The plugin is the same durability by a different route: install it once, and the rules arrive as a skill and the MCP server as part of the plugin. Nothing in the user's project is touched, and updating the plugin is how it stays current.

Both routes are supported. The plugin suits someone who reviews across several projects; tailr init suits a project that wants Tailr committed as part of its own setup.

Keeping it honest

The rules in skills/review/SKILL.md are generated from src/setup/rules.js, the same source tailr init writes from. Every catalog's version, and the Gemini copy of the skills, is stamped from package.json.

node scripts/plugin.js --check    is every catalog in step with the source?
node scripts/plugin.js --sync     put them back in step

--check runs in the test suite, so CI fails on drift and so does the release workflow at the tag. --sync runs as part of npm version, so cutting a release stamps every catalog without anyone having to remember it.