git:20261022.221576e to git:20261022.8d40669

1 added, 1 removed. Audit A to A.

# AGENTS.md

Guidance for coding agents working in this repository. Humans: see [CONTRIBUTING.md](./CONTRIBUTING.md), which holds the rules every skill follows.

## Layout

- `skills/<name>/SKILL.md`: the skills. Each folder also has `agents/openai.yaml`.

- `.claude-plugin/`, `.codex-plugin/`, `.cursor-plugin/`, `gemini-extension.json`: plugin manifests. `mcp.json` is the shared MCP server entry.

- - `server.json`: the MCP Registry entry `dev.jam/jam`. Pushing a `v*` tag publishes it, and a published version is permanent. Never push a tag unless asked.

+ - `server.json`: the MCP Registry entry `dev.jam/jam`. Its `version` is the Jam MCP server's, not the plugin's. Pushing an `mcp-v*` tag publishes it, and a published version is permanent. Never push a tag unless asked.

- `rules/`: a Cursor rule that guards the Jam MCP delete tools.

- `docs/plugin.md`: plugin sign-in, what the MCP server can do, permissions. The tool list itself lives at jam.dev/mcp; do not copy it here.

- `scripts/check.mjs`: repo checks, run by CI.

## Invariants

Run `node scripts/check.mjs` after any change. It prints each rule it finds broken, with the file.

Also:

- When you add, rename or remove a skill, update its README card, the CHANGELOG, and any skill that names it.

- `jam-proof` and `jam-annotate` are also bundled in the Jam CLI. This repo is the source; tell the user the bundle needs a resync when you change them.

- Only name `jam` CLI commands and MCP tools that ship today. Check with `jam agent-context`.

- Skills are public. No internal paths, repo names, staging hosts or customer data.