commit-pattern-analyzer

name	Commit Pattern Analyzer
description	Analyze recent git commits in a target project to surface patterns, identify improvement opportunities, and generate recommendations weighted by source authority and recency. Trigger when user says "analyze commits", "review my recent work", "what patterns do you see", or when evaluating a project for best-practice alignment.
allowed-tools	Read, Grep, Glob, Bash

Commit Pattern Analyzer

Surface actionable recommendations by analyzing how work is actually being done, not just what the code looks like.

Trigger Conditions

Activate when:

• User says "analyze commits", "review my recent work", "what patterns do you see"
• Evaluating a project for best-practice alignment
• After running ONE-LINE-PROMPT.md evaluation
• User asks "how can I improve my workflow"

Skip when:

• User is asking about a specific bug or feature (use normal tools)
• Repository has fewer than 10 commits
• User explicitly asks for code review, not workflow review

Workflow

Step 1: Gather Commit Data

# In the target project (default 90 days, use 365 for low-activity repos):
git log --oneline --since="90 days ago" --format="%H|%s|%an|%ad" --date=short
git log --since="90 days ago" --stat --format="" | sort | uniq -c | sort -rn | head -20
git log --since="90 days ago" --format="%s" | grep -i "Co-Authored-By" | wc -l

Step 1.5: Inspect Harness Structure

# Check for Claude Code infrastructure:
ls -la CLAUDE.md .claude/CLAUDE.md 2>/dev/null
wc -l CLAUDE.md .claude/CLAUDE.md 2>/dev/null
ls -la .claude/settings.json 2>/dev/null
ls .claude/rules/*.md 2>/dev/null | wc -l
ls .claude/hooks/ 2>/dev/null
ls -d .claude/skills/*/ 2>/dev/null | wc -l
ls -d .claude/commands/*/ 2>/dev/null | wc -l

Score 0-6 (one point per component present):

• CLAUDE.md (root or .claude/) — presence + line count
• .claude/settings.json — permissions configured?
• .claude/rules/*.md — domain rules exist?
• .claude/hooks/ — lifecycle hooks?
• .claude/skills/ — reusable workflows?
• .claude/commands/ — slash commands?

Step 2: Identify Patterns

Analyze the commit data for:

• File types changed most: Which areas of code get the most attention?
• Commit message patterns: Are messages descriptive? Conventional commits?
• AI co-authoring ratio: What % of commits are Co-Authored-By AI?
• Commit frequency: Bursts vs steady? Time of day?
• Refactoring vs feature work ratio: How much is new vs maintenance?
• Test co-location: Are tests committed alongside features?

Step 3: Cross-Reference Against Analysis Docs

Map observed patterns to the 26 analysis documents in analysis/:

• High AI co-authoring → agent-driven-development.md patterns
• CLAUDE.md changes → claude-md-progressive-disclosure.md
• Skill/workflow changes → mcp-vs-skills-economics.md
• Test patterns → agent-evaluation.md
• Large commits → harness-engineering.md (one feature at a time)

Step 4: Weight Recommendations

Apply source-authority-matrix weighting:

• Recommendations backed by Tier 5 (Foundational) sources → high priority
• Recommendations backed by Tier 3 (Practitioner) sources → medium priority
• Recommendations backed by Tier 2 (Commentator) sources → note, don't push

Step 5: Output

## Commit Pattern Analysis: [project-name]

**Period**: [date range]
**Commits analyzed**: [count]
**AI co-authoring rate**: [X%]

### Patterns Observed
1. [pattern] — [frequency/evidence]
2. [pattern] — [frequency/evidence]

### Recommendations (by priority)

#### High Priority (Foundational source backing)
- [recommendation] — Source: [name] (Authority [tier], Weight [score])

#### Medium Priority (Practitioner source backing)
- [recommendation] — Source: [name] (Authority [tier], Weight [score])

#### Worth Noting (Commentator source backing)
- [observation] — Source: [name] (Authority [tier], Weight [score])

Don't

• Recommend changes without checking actual commit evidence
• Weight YouTube-sourced recommendations equally with Anthropic engineering blog
• Produce recommendations longer than the analysis — be concise
• Criticize workflow choices without understanding project constraints