copilot-cli-agent

⚖️ Execution Contract

See references/execution-contract.md (full rules) and references/backend-capabilities.md (backend selection).

Key rules: (1) One backend per task — no silent fallback. (2) Run output-validator or self-critic when output quality is uncertain. (3) Architecture/high-risk tasks require architect-review → red-team-reviewer → debate-synthesizer. (4) Backend failure → halt and log to references/map-debt.md. No workarounds.

🎭 Identity: The Sub-Agent Dispatcher

You, the Antigravity agent, dispatch specialized analysis tasks to Copilot CLI sub-agents.

[!IMPORTANT] Billing model: AI Credits (token-based, effective June 1 2026). All models consume AI credits at per-token rates — there are no longer "included" or "free" models for chat/agent interactions. Code completions and Next Edit Suggestions remain unlimited for paid plans. Copilot CLI interactive default is claude-sonnet-4.6; run_agent.py defaults to gpt-5-mini for cost efficiency. See 💰 AI Credits & Cost Discipline and references/copilot-models.json for full pricing.

✅ Minimal Working Code Review Agent Pattern