Add model-aware Copilot delegation policy and repository committer agent

[Copilot]

Goal

Claude/Codex already enforced lead-authors/delegate-executes, but Copilot CLI and GitHub Copilot did not. This PR adds the same delegation contract for Copilot surfaces, including explicit model-handling policy to avoid undocumented escalation behavior.

Scope

• Shared Copilot policy (cross-surface)

  • Added .github/copilot-instructions.md as the common instruction source for Copilot CLI + GitHub Copilot.
  • Defines bounded delegation categories: commit, read-only recon, independent lint/test lanes, boilerplate drafts, research summaries, and mechanical GitHub chores.

• Repository custom committer agent

  • Added .github/agents/committer.md with a strict mechanical contract:
    • requires exact lead-authored commit message + exact file list
    • cross-checks against git status
    • stages only named paths
    • commits verbatim (no trailers/rewording)
    • never bypasses hooks, never fixes hook failures, never pushes
  • Marked manual-only (disable-model-invocation: true).

• Model policy + documentation alignment

  • Updated AGENTS.md mechanism map and delegation section to include Copilot instruction/agent layers.
  • Updated .github/README.md with Copilot customization architecture and a CLI/GitHub validation checklist.
  • Updated setup/MIGRATION.md concept map + migration step for adopting Copilot instruction and agent files, including model pinning/fallback guidance.

• Example (agent profile shape)

  ---
  name: committer
  tools: [execute, read, search]
  disable-model-invocation: true
  ---

Risk & rollback

Low risk: changes are limited to documentation and agent-profile configuration; no runtime application logic or CI workflow behavior was modified.
Rollback is a straight revert of the five touched paths (AGENTS.md, .github/README.md, setup/MIGRATION.md, .github/copilot-instructions.md, .github/agents/committer.md).

Evidence

• The repo now contains a single shared Copilot instruction source plus a bounded committer profile under .github/agents/, matching the lead-authors/delegate-executes architecture used elsewhere.
• The committer profile enforces constrained tooling and manual invocation, and intentionally leaves model unpinned to inherit caller/default behavior until cross-surface pinning behavior is explicitly validated and documented.
• Migration and .github docs now include explicit cross-surface validation criteria (discovery, invocation, tool restrictions, and contract behavior), making verification reproducible for downstream adopters.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 51484dabdf

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use the required .agent.md suffix

Because this profile is named committer.md, Copilot will not discover it when the documented repository-agent filename pattern is *.agent.md; the official custom-agent documentation specifies that agent profile filenames must end in .agent.md (GitHub: custom agents). Consequently, the new committer agent cannot appear in the CLI or repository custom-agent picker as promised. Rename it to committer.agent.md and update the references to that path.

Useful? React with 👍 / 👎.