code-feedback

Code feedback

Give code-review feedback at PR level or across the whole codebase. You coordinate review subagents and synthesize; you never approve or merge.

1. Ask the user

Use AskUserQuestion to settle:

• Scope — a specific PR (by number) / all open pipeline PRs / the whole codebase.
• Focus — correctness, security, performance, simplification, tests/TDD adherence, or all.

2a. PR scope

• Read the diff (gh pr diff <pr>) and the PR/issue context. Fan out one review subagent per focus dimension for depth; each returns concrete findings (file, line, severity, why, suggested fix).
• Verify findings before posting (drop the speculative ones). Post one review via the native API: bash "$S/pr-review.sh" <pr> --event COMMENT --body-file <summary.md> [--comments-file <inline.json>] where inline.json is [{ "path","line","body" }, ...]. Use COMMENT (or REQUEST_CHANGES); never approve or merge — feedback only.

2b. Whole-codebase scope

• Fan out parallel Explore/review subagents across subsystems (correctness, security, performance, dead code, test coverage). Synthesize a findings report grouped by severity, using CONTEXT.md vocabulary and respecting docs/adr/.
• If the user opts in, open issues for actionable findings: gh issue create ... --label maestro:ready-for-agent --assignee @me (add maestro:tech-debt for debt) so /maestro:ship or /maestro:auto can fix them; set dependencies where findings build on each other.

3. Report

Summarize what you reviewed, the review(s) posted, and any issues opened. Locate scripts first: source .maestro/config.sh; S="${MAESTRO_SCRIPTS:-$(pwd)/.maestro/scripts}".