Commit Analyzer

Purpose

Analyzes git commit history to surface patterns, recurring issues, and extractable knowledge. Designed to feed insights into /claudeception for continuous learning and skill generation.

When to Use

• After a debugging session to identify what went wrong and why
• Before refactoring to understand the evolution of problematic code
• When noticing repeated fixes for similar issues
• During retrospectives to extract learnings
• When the user asks to “analyze commits” or “review commit history”

Analysis Process

Step 1: Gather Commit Data

# Get recent commits with stats
git log --oneline --stat -n 60

# Get commits with full messages for categorization
git log --pretty=format:"%h|%s|%b" -n 60

# Get commits touching specific files (for focused analysis)
git log --oneline --follow -- path/to/file

Step 2: Categorize Commits

Analyze commit messages and classify into categories:

Step 3: Identify Patterns

Look for these anti-patterns:

1. Repeated Fixes: Multiple commits fixing the same component/file

   • Signal: Same filename in 3+ fix commits
   • Extract: Root cause analysis needed

2. Fix-Feature Ratio: High ratio indicates technical debt

   • 40% fixes = Concerning

   • 60% fixes = Critical

3. Churn Files: Files with many modifications

   • Signal: >10 commits touching same file
   • Extract: Might need architectural review

4. Cascading Fixes: Fix followed by “fix fix” or revert

   • Signal: Commits referencing previous commits
   • Extract: Testing/review process gap

5. Pattern Clusters: Similar fixes across different files

   • Signal: Same keywords (e.g., “env var”, “null check”) in multiple fixes
   • Extract: Systematic issue, needs pattern fix

Step 4: Generate Root Cause Analysis

For each pattern found, generate:

## Root Cause: [Name]

**Evidence:**
- Commit A: [hash] [message]
- Commit B: [hash] [message]
- Commit C: [hash] [message]

**Pattern:** [Description of what's happening]

**Root Cause:** [Why this keeps happening]

**Fix:** [What would prevent this class of bugs]

**Skill Extraction:** [If this should become a Claude Code skill]

Step 5: Feed to Claudeception

After analysis, automatically invoke /claudeception with:

Based on the commit history analysis:

## Recurring Issues Found

[List of patterns with evidence]

## Proposed Skills to Extract

1. **[skill-name]** - [Description of when to use and what it solves]
   - Trigger: [When this skill should activate]
   - Solution: [What the skill should do]

Please review and create appropriate skills for the valuable learnings.

Example Output

# Commit Analysis Report: clawdbot-railway

## Summary
- Total commits analyzed: 60
- Bug fixes: 26 (43%)  <-- WARNING: High fix ratio
- Features: 18 (30%)
- Refactors: 8 (13%)
- Other: 8 (14%)

## Recurring Patterns

### Pattern 1: Module-Level Env Var Reads (6 commits)
**Commits:**
- ee510f6: Fix env var timing in cron tracker
- a1b2c3d: Fix SUPABASE_URL undefined in health check
- x4y5z6a: Fix module-level constant capture

**Root Cause:** In serverless environments, module-level `const X = process.env.X`
can capture `undefined` before env vars are initialized during cold start.

**Skill to Extract:** `vercel-serverless-env-vars`

### Pattern 2: Markdown Escaping (4 commits)
**Commits:**
- 52a110c: Fix markdown escaping in telegram
- 8c9be65: Another markdown escape fix
- 2f89a9d: Escape backslash first

**Root Cause:** Telegram MarkdownV2 requires specific escape order (backslash first)
and 3 different implementations exist.

**Skill to Extract:** `telegram-markdownv2-escaping`

## Recommended Actions

1. Create skill: `vercel-serverless-env-vars` (already exists - verify usage)
2. Create skill: `telegram-markdownv2-escaping`
3. Consolidate 3 escapeMarkdown implementations into one
4. Add pre-commit TypeScript check

## Feeding to /claudeception...
[Auto-invokes claudeception with findings]

Integration with Claudeception

This skill should automatically call /claudeception when:

• Analysis finds 2+ recurring patterns
• A pattern has 3+ related commits
• A root cause is identified that isn’t already captured in a skill

The handoff format:

/claudeception

Session learning from commit analysis:

1. **Pattern Name**: [Description]
   - Evidence: [Commit hashes]
   - Root cause: [Why it happens]
   - Solution: [How to prevent]

Please extract any valuable skills from these findings.

Command Aliases

• /commit-analyzer – Run full analysis
• /analyze-commits – Alias
• /commit-patterns – Alias
• “analyze my commit history” – Natural language trigger
• “what patterns do you see in recent commits” – Natural language trigger

Notes

• Focus on patterns, not individual commits
• Weight security and fix commits higher in analysis
• Always cross-reference with existing skills to avoid duplication
• Consider project-specific vs universal patterns
• Invoke claudeception only for genuinely useful learnings

References

• Claudeception skill: ~/.claude/skills/claudeception/SKILL.md
• Continuous learning system: Extracts reusable knowledge from sessions