CHAP 0.3.1: create-chap-app, and a gate for every commit a coding agent makes

The protocol packages move together to 0.3.1 and behave as in 0.3.0; the
release brings create-chap-app to npm. chap-analytics 0.2.2, on its own
track, reads the store of the TypeScript coordinator, which the Node templates
run.

npx create-chap-app my-project

Added

• create-chap-app. npx create-chap-app generates a project that runs
  against the developer's own agent and data: one process owns the SQLite store
  and serves a review desk and POST /chap, the agent is a separate process
  attached through an MCP client, a framework bridge or the HTTP endpoint, the
  profiles are one setting in chap.config.json, and diff-profiles runs the
  project's workload under two profile sets and prints each call's outcome under
  both. Five templates: the code gate below; an MCP gate for Claude Desktop,
  Cursor and Claude Code; a support desk and an outbound approval gate in
  Python; and the Handbook's production set with signed calls, a token
  presented at join verified against its issuer, the chain on, and a doctor.
  The design.
• The code gate. A template that puts every commit a coding agent makes
  behind a person's decision. The agent's change is proposed as a patch and
  reviewed as a diff, and only an approved change is committed, with four
  trailers (the model that drafted it, read from the agent's own harness, each
  approving reviewer by name and email, the committer's sign-off, and the
  approval it carries) and a note under refs/notes/chap holding the agent's
  signed proposal and each reviewer's signed decision. An agent that commits on
  its own has its branch reviewed commit by commit before it is pushed. One
  command runs the review loop: it starts the gate, opens the review in the
  browser, waits, and hands the agent the reviewer's decision as a prompt,
  with their comments on lines. verify.mjs checks a range of commits against
  the team's trust policy, and a workflow runs it on every pull request.
• The desk, rebuilt. Every template's desk shows the artefact by its shape,
  lets a reviewer request changes as well as approve, edit or reject, reads the
  chain as activity, and counts in an Insights view what reviewers accept, edit
  and send back, by model and, for code, by file. Its text grows with the
  window, from a phone to a large monitor.
• Analytics in every project. analytics.py writes the chap-analytics
  report, the evaluation cases and a refinement page from the project's store,
  and the desk links them.
• The profile explorer.
  docs/profile-explorer.md
  shows, for each profile, the calls whose outcome changes when the profile is
  added.
• The playground drafts with any model, or none. It reads
  ANTHROPIC_API_KEY, OPENAI_API_KEY or OLLAMA_URL, and drafts with a
  scripted agent when none is set.

Changed

• @brightbeamai/chap-coordinator-mcp needs @modelcontextprotocol/sdk
  1.32.1 or later.

chap-analytics 0.2.2

• A TypeScript coordinator's store. from_sqlite and from_json read a
  workspace snapshot the TypeScript coordinator wrote; both coordinators'
  stores project to the same tables.

The full entry is in the
changelog.