/commit
Use when ready to commit changes. Runs CodeRabbit review first, then commits if review passes. Supports Ralph mode for atomic commit + criterion marking. Covers commit, ralph commit, atomic commit. NOT for: pushing or creating PRs (use pr-loop).
$ golems-cli skills install commitUpdated 2 weeks ago
Runs CodeRabbit review on staged changes, then commits if approved.
Flow
- Check for staged changes
- Run CodeRabbit review (
cr review --plainfor headless/Claude compatibility) - Show review results
- If review passes → prompt for commit message → commit
- If review fails → show issues → ask user if they want to proceed anyway
Usage
# Stage your changes first
git add <files>
# Standard commit
/commit
# Ralph mode — atomic commit + mark story criterion
/commit --story=US-106 --message="feat: US-106 description"What This Skill Does
When invoked, Claude will:
- Check staged changes: Run
git diff --staged --statto show what's staged - Run CodeRabbit: Execute
cr review --plain(headless mode, works from Claude) - Evaluate results:
- If CR passes (no critical issues): proceed to commit
- If CR fails: show issues and ask user for decision
- Commit: Generate commit message based on changes, commit with co-author
Full SKILL.md source — includes LLM directives, anti-patterns, and technical instructions stripped from the Overview tab.
Runs CodeRabbit review on staged changes, then commits if approved.
Flow
- Check for staged changes
- Run CodeRabbit review (
cr review --plainfor headless/Claude compatibility) - Show review results
- If review passes → prompt for commit message → commit
- If review fails → show issues → ask user if they want to proceed anyway
Usage
# Stage your changes first
git add <files>
# Standard commit
/commit
# Ralph mode — atomic commit + mark story criterion
/commit --story=US-106 --message="feat: US-106 description"What This Skill Does
When invoked, Claude will:
- Check staged changes: Run
git diff --staged --statto show what's staged - Run CodeRabbit: Execute
cr review --plain(headless mode, works from Claude) - Evaluate results:
- If CR passes (no critical issues): proceed to commit
- If CR fails: show issues and ask user for decision
- Commit: Generate commit message based on changes, commit with co-author
Ralph Mode (Story Commits)
When --story is provided, this skill does atomic commit + criterion marking:
- Stages files (specified or auto-detected)
- Commits (pre-commit hook runs all tests)
- If commit succeeds → marks the commit criterion as checked in story JSON
- If commit fails → neither happens, reports failure
Ralph Mode Flags
| Flag | Description |
|---|---|
--story=ID | Story ID (e.g., US-106, BUG-028) — triggers Ralph mode |
--message=MSG | Commit message (required in Ralph mode) |
--files=PATHS | Files to stage (default: prd-json/ + modified files) |
--dry-run | Show what would happen without doing it |
Why Ralph Mode Exists
- Atomic: Either both happen (commit + check) or neither
- No double-testing: Uses pre-commit hook, doesn't run tests twice
- Clean failure: If tests fail, criterion stays unchecked for retry
Requirements
crCLI installed (CodeRabbit)- Changes staged with
git add
After Committing
Committing is ONE step in the full workflow. After commit:
- Push →
git push -u origin <branch> - PR + Review + Merge →
/pr-loop(the full loop)
Do NOT stop at commit. The mission is MERGED, not committed.
Best Pass Rate
100%
Claude Sonnet
Assertions
13
2 models tested
Avg Cost / Run
$0.0133
across models
Fastest (p50)
1.6s
Claude Sonnet
Behavior Evals
Phase 2 baseline — skill quality on ClaudeBehavior Baseline
Adapter Evals
Phase 2C — cross-AI portabilityAdapter Portability
| Assertion | Claude Sonnet | Codex (GPT-5.4) | Consensus |
|---|---|---|---|
| staged-diff-is-inspected-first | 2/2 | ||
| coderabbit-gates-the-commit | 2/2 | ||
| commit-message-follows-review-pass | 2/2 | ||
| coauthor-trailer-is-included | 2/2 | ||
| review-findings-are-surfaced | 2/2 | ||
| automatic-commit-is-blocked-on-fail | 2/2 | ||
| override-or-fix-choice-is-explicit | 2/2 | ||
| does-not-claim-commit-success | — | 1/2 | |
| no-staged-diff-is-detected | 2/2 | ||
| review-is-not-run-without-staged-changes | — | 1/2 | |
| user-is-told-to-stage-files-first | — | 1/2 | |
| standard-commit-route-points-to-codex | — | 1/2 | |
| ralph-mode-is-marked-unavailable | — | 1/2 |
Token Usage
Cost per Run
| Model | Input Tokens | Output Tokens | Cost / Run | Cost / 1K Runs |
|---|---|---|---|---|
| Claude Sonnet | 1,500 | 500 | $0.0068 | $6.80 |
| Codex (GPT-5.4) | 1,800 | 650 | $0.0198 | $19.80 |
Response Time (p50)
Response Time (p95)
| Model | p50 | p95 | Overhead |
|---|---|---|---|
| Claude Sonnet | 1.6s | 2.8s | +75% |
| Codex (GPT-5.4) | 2.4s | 4.2s | +75% |
Last evaluated: 2026-03-12 · Real Phase 2 evals · behavior (Claude) + adapter (1 CLI)
Changelog entries are derived from eval runs and skill version updates. Full cascading changelog (Phase 4D) coming soon.
Best Pass Rate
100%
Assertions
13
Models Tested
2
Evals Run
3
- +Initial release to Golems skill library
- +10 assertions across 3 eval scenarios
- +Eval fixtures included