Agent setup

You don't carry the methodology in your head, and you don't paste the canon into a chat. The agent learns the rules from the tool; you bring the judgment. Two pieces make that automatic.

1. The repo carries its own brief

Drop a CLAUDE.md at the repo root. Claude Code reads it on open, so any agent you start there is briefed before you've said a word. A ready template lives atdocs/conventions/specline-CLAUDE.md; it tells the agent it's a Specline repo, how to shape and migrate plans with you, and to reach for Specline tools.

2. Specline is sitting right there as a tool

With the MCP server registered at user scope, the agent discovers specline_spec / specline_rules / specline_checkin every repo. The brief tells it where it is; the tools let it act. It onboards itself.

What a day looks like

1. Open a fresh agent chat in the repo (with CLAUDE.md present).
2. Say it in your own words:
   you →

   Let's shape a new feature: <your idea>.
   — or —
   Migrate this plan into Specline with me: <the file>.

3. The agent pulls the canon (specline_spec), asks you the shaping questions, and writes the spec folder in docs/drafts/ — checking itself with specline_check as it goes (only parse-level integrity runs against a draft).
4. You read the result in one sitting and approve it — moving it to docs/specs/ — or send it back. That's the approve gate — your call, not the machine's.

You don't need Claude Code for this

Any agent with shell access does the same: have it run specline spec to load the canon and specline check --format json to check its work. MCP just makes the discovery automatic.