ai-reviewed-commit

name	ai-reviewed-commit
description	Use when: committing with pre-commit review, authorship-aware fix-retry
allowed-tools	Bash, Read, Grep, Glob, Edit
triggers	["ai-reviewed-commit","/ai-reviewed-commit","reviewed commit","review before commit"]

AI-Reviewed Commit: Pre-Commit Review Loop

Role Detection (READ THIS FIRST -- MANDATORY)

Check your initial prompt. If it contains the literal string "Use the ai-reviewed-commit skill", then:

YOU ARE THE EXECUTOR. Follow these rules:

1. SKIP the "Execution Protocol" section below -- it is for the orchestrator, not you.
2. Read the pattern at ../../PATTERN.md relative to this SKILL.md, and follow it step by step.
3. ABSOLUTE PROHIBITION: Do NOT run csa run, csa review, csa debate, or ANY csa command. You must perform the work DIRECTLY. Running any csa command causes infinite recursion.

Only if you are the main agent (Claude Code / human user):

• You are the orchestrator. Follow the "Execution Protocol" steps below.

---

Purpose

Ensure every commit is reviewed by an independent model before creation. Uses authorship-aware strategy: self-authored code gets adversarial debate review (csa debate), while other-authored code gets standard csa review --diff --allow-fallback. The compiled weave workflow is a single-pass mechanical gate, while the orchestrator is responsible for driving additional rounds under the contract below. Includes an automated fix-and-retry loop with a 3-round hard cap, mandatory AGENTS.md compliance checking, and Conventional Commits message generation.

Execution Protocol (ORCHESTRATOR ONLY)

Prerequisites

• Changes must be staged: git diff --staged shows content

Quick Start

csa run --sa-mode true --skill ai-reviewed-commit "Review and commit the staged changes"

SA Mode Propagation (MANDATORY)

When operating under SA mode (e.g., dispatched by /sa or any autonomous workflow), ALL csa invocations MUST include --sa-mode true. This includes csa run, csa review, csa debate, and any other execution commands. Omitting --sa-mode at root depth causes a hard error; passing false when the caller is in SA mode breaks prompt-guard propagation.

Step-by-Step

1. Stage changes: git add the target files.
2. Size check: Run git diff --stat --staged. If >= 500 lines, consider splitting the commit.
3. Authorship detection: Determine if staged code is self-authored (generated in this session) or by another tool/human.
4. Review:
   • Self-authored: csa debate "Review my staged changes for correctness, security, and test gaps. Run 'git diff --staged' yourself."
   • Other-authored: csa review --diff --allow-fallback
5. Fix loop (if issues found): Dispatch sub-agent to fix issues, re-stage, re-review. Preserve original code intent -- do NOT delete code to silence warnings. Fix-and-retry up to 3 rounds (hard cap). After round 3, if review still reports non-false-positive P0/P1 findings, STOP and ask the user whether to continue. Exception: if the user's prior prompt explicitly authorized unbounded looping (e.g., "loop until clean", "keep fixing until review passes"), continue without asking. Also continue without asking if all round-3 findings are false positives per orchestrator judgement.
   • The weave workflow itself remains single-pass and only provides the deterministic halt gate plus one review/fix/re-review pass.
   • Rounds 2 and 3 are driven by the orchestrator re-invoking the contract, not by native weave WHILE/GOTO.
   • The hard cap and exception rules above are binding on the orchestrator even when weave cannot mechanically express the full loop.
   • Multi-finding optimization: When a fix round has 2+ findings in different files, use the parallel-fix pattern (parallel RECON / serial EDIT) instead of a single-employee fix. See patterns/parallel-fix/skills/parallel-fix/SKILL.md.
6. AGENTS.md compliance: Discover AGENTS.md chain for each staged file. Check every applicable rule. If the staged diff or generated commit body lists concrete Timing/Race Scenarios, verify that matching regression tests exist and are named under Regression Tests Added; missing or mismatched tests are a blocking failure. Zero unchecked items before proceeding.
7. Generate commit message: csa run "Run 'git diff --staged' and generate a Conventional Commits message" (tier-1).
8. Commit: git commit -m "${COMMIT_MSG}".

Example Usage

Command	Effect
/ai-reviewed-commit	Review staged changes and commit
/ai-reviewed-commit files=src/lib.rs	Stage, review, and commit specific file

Integration

• Uses: csa-review (for other-authored code), debate (for self-authored code)
• Used by: commit (Step 8 pre-commit review)
• Produces: A single reviewed commit with Conventional Commits message

Done Criteria

1. Changes staged and diff size checked.
2. Authorship-appropriate review strategy selected and executed.
3. All review issues fixed, or the workflow stopped at the 3-round hard cap and requested user direction before continuing.
4. AGENTS.md compliance checklist completed with zero unchecked items.
5. Conventional Commits message generated via CSA.
6. Commit created successfully.
7. git status confirms commit was created.