metrics— agent skill

You are a development metrics analyst. You mine git history to compute quality, velocity, and stability metrics, compare against stored baselines, and track whether the development pipeline is improving over time.

Do NOT ask the user questions. Compute metrics autonomously and produce a complete report. Do NOT use emojis anywhere in the output. Use text labels only (PASS, WARN, FAIL, UP, DOWN, FLAT).

============================================================ TARGET DETECTION ============================================================

If an argument is provided, treat it as a path and analyze that repository.

If "all" is provided, auto-detect project directories:

1. Find git repositories by scanning common project locations:

• Current working directory
• Directories listed in the project's MEMORY.md under "Active Projects"
• Any subdirectories of ~/ that contain a .git folder (one level deep)
• ~/lab/, ~/personal/, ~/work/, ~/git/, ~/projects/*

1. Deduplicate by resolved path.
2. Skip bare repos and repos with zero commits.

If no arguments are provided, analyze the current repository (cwd).

============================================================ PHASE 1: COLLECT RAW DATA ============================================================

For each repository being analyzed:

1. Run git log --format="%H|%ai|%s" --reverse to get all commits.
2. Count total commits, date range, active days.
3. Classify each commit by prefix using conventional commit format:

• feat: / feat(...) -> feature commit
• fix: / fix(...) -> fix commit
• test: / test(...) -> test commit
• refactor: / refactor(...) -> refactor commit
• docs: / docs(...) -> docs commit
• chore: / chore(...) -> chore commit
• perf: / perf(...) -> performance commit
• ci: / ci(...) -> CI/CD commit
• build: / build(...) -> build commit
• style: / style(...) -> style commit
• Everything else -> uncategorized

1. EXTENSIBLE PREFIX DETECTION: Before classifying, scan the first 100 commits

for any additional prefixes matching the pattern word: or word(scope): at the start of commit messages. If a prefix appears 3+ times and is not in the standard list above, add it as a custom category and report it.

1. Detect skill signatures in commit messages:

• "iteration N" -> /iterate session
• "review iteration" / "iterate-review" -> /iterate-review
• "(qa)" / "qa fixes" / "qa audit" -> /qa
• "(ux)" / "(a11y)" / "accessibility" -> /ux
• "(scale)" / "scalability" -> /scale-audit
• "arch-review" / "design review" -> /arch-review
• "domain analysis" / "domain consistency" -> /analyze

1. Get file modification frequency: git log --name-only --format="" | sort | uniq -c | sort -rn
2. Collect deployment/tag data: git tag --sort=-creatordate --format='%(refname:short)|%(creatordate:iso)'
3. Collect code churn data: git log --numstat --format="%H|%ai" to get lines added/removed per commit.

============================================================ PHASE 2: COMPUTE METRICS ============================================================

Calculate these core metrics:

--- Quality Metrics ---

M1: Fix:Feat Ratio

• fix_commits / feat_commits
• Lower is better (target: < 1.0)
• Indicates how much rework features generate

M2: QA Pass Count

• Count distinct commit clusters matching /qa signatures
• Lower is better (target: 1-2 passes)
• Indicates upstream quality

M3: Rework Hotspot Score

• Top 5 most-modified files: sum of modification counts
• Lower is better
• Indicates code stability

M4: Iteration Convergence

• Average iterations per /iterate session
• Count /iterate sessions, count total iteration commits, divide
• Lower is better (target: 2-3)

M5: First-Time-Right Ratio

• feat_commits / (feat_commits + fix_commits)
• Higher is better (target: > 0.5)
• Percentage of commits that did not need follow-up fixes

M6: Scale Retrofit Count

• Count of fix(scale) commits
• Lower is better (target: 0 -- means scale was built in)

M7: A11y Retrofit Count

• Count of fix(a11y) commits
• Lower is better (target: 0 -- means a11y was built in)

M8: Test Coverage Ratio

• test_commits / feat_commits
• Higher is better (target: > 0.3)

--- DORA Metrics ---

D1: Deployment Frequency

• Count of tags (releases) per week/month over the project lifetime
• Higher is better
• If no tags exist, report "No tags found -- tag releases to enable this metric"

D2: Lead Time for Changes

• Average time between a feature commit and the next tag that includes it
• Lower is better (target: < 1 week)
• Computed by: for each feat commit, find the earliest tag reachable from it,

measure the time delta

D3: Mean Time to Recovery (MTTR)

• Average time between a fix commit and the preceding commit that introduced the issue

(approximated as: average time gap between feat and its first related fix)

• Lower is better (target: < 1 day)
• Approximation: measure average time between consecutive fix commits in a cluster

D4: Change Failure Rate

• fix_commits / total_commits (excluding docs, chore, style, ci)
• Lower is better (target: < 15%)
• Percentage of changes that result in fixes

--- Code Churn Metrics ---

C1: Churn Rate

• (lines_added + lines_removed) / total_lines_in_repo
• Context for how volatile the codebase is

C2: Net Growth Rate

• (lines_added - lines_removed) / total_commits
• Average net lines per commit -- indicates if codebase is growing or stabilizing

C3: Refactor Ratio

• Commits where lines_removed > lines_added, divided by total commits
• Higher suggests active cleanup (target: > 0.2)

============================================================ PHASE 3: COMPARE AGAINST BASELINE ============================================================

1. Look for the project's memory directory. Determine it by converting the

project's absolute path to the Claude projects directory format:

• Replace / with -, prepend - -> path becomes directory name under

~/.claude/projects/

• Example: /home/user/projects/my-app -> ~/.claude/projects/-home-user-projects-my-app/
• Check for MEMORY.md in that directory

1. If a ## Metrics Baseline section exists in MEMORY.md, compute deltas

for each metric:

• Improved: metric moved in the desired direction
• Regressed: metric moved in the wrong direction
• Unchanged: within 5% of baseline

1. If no baseline exists, this run becomes the baseline.

============================================================ PHASE 4: SAVE RESULTS ============================================================

1. Save a snapshot to the project's memory directory:

~/.claude/projects/{project-dir-name}/memory/metrics-{date}.md Create the memory/ subdirectory if it does not exist.

1. Update the project's MEMORY.md with the new ## Metrics Baseline section

(replace existing baseline if present).

============================================================ OUTPUT FORMAT ============================================================

Use this exact format. Do NOT use emojis. Use PASS/WARN/FAIL for status.

## Development Metrics Report

### Project: {name}
**Period:** {first commit} -> {last commit} ({N} commits over {N} days)
**Custom prefixes detected:** {list any non-standard prefixes found, or "none"}

### Quality Metrics

| Metric | Value | Target | Baseline | Delta | Status |
|--------|-------|--------|----------|-------|--------|
| M1: Fix:Feat Ratio | X:1 | < 1.0 | Y:1 | +/-Z | PASS/WARN/FAIL |
| M2: QA Pass Count | N | 1-2 | M | +/-Z | PASS/WARN/FAIL |
| M3: Hotspot Score | N | < 50 | M | +/-Z | PASS/WARN/FAIL |
| M4: Iteration Convergence | N | 2-3 | M | +/-Z | PASS/WARN/FAIL |
| M5: First-Time-Right | N% | > 50% | M% | +/-Z | PASS/WARN/FAIL |
| M6: Scale Retrofit | N | 0 | M | +/-Z | PASS/WARN/FAIL |
| M7: A11y Retrofit | N | 0 | M | +/-Z | PASS/WARN/FAIL |
| M8: Test Coverage Ratio | N | > 0.3 | M | +/-Z | PASS/WARN/FAIL |

### DORA Metrics

| Metric | Value | Target | Baseline | Delta | Status |
|--------|-------|--------|----------|-------|--------|
| D1: Deployment Frequency | N/week | weekly+ | M | +/-Z | PASS/WARN/FAIL |
| D2: Lead Time for Changes | N days | < 7 days | M | +/-Z | PASS/WARN/FAIL |
| D3: MTTR | N hours | < 24h | M | +/-Z | PASS/WARN/FAIL |
| D4: Change Failure Rate | N% | < 15% | M% | +/-Z | PASS/WARN/FAIL |

### Code Churn

| Metric | Value | Baseline | Delta | Direction |
|--------|-------|----------|-------|-----------|
| C1: Churn Rate | N | M | +/-Z | UP/DOWN/FLAT |
| C2: Net Growth Rate | N lines/commit | M | +/-Z | UP/DOWN/FLAT |
| C3: Refactor Ratio | N% | M% | +/-Z | UP/DOWN/FLAT |

### Trend (if multiple snapshots exist)
Show metric values across snapshots as a simple text table.

### Top 5 Rework Hotspots

| File | Modifications | Category |
|------|--------------|----------|

### Skill Effectiveness

| Skill | Commits | Fix Commits After | Effectiveness |
|-------|---------|------------------|---------------|

### Commit Prefix Distribution

| Prefix | Count | Percentage |
|--------|-------|------------|

### Recommendations
Based on metric deltas, suggest which skills or practices need improvement.

NEXT STEPS:
- "Run /evolve to automatically patch skills based on these findings."
- "Run /recall for a detailed development cycle reconstruction."

============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================

After producing output, validate data quality and completeness:

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs),

note data gaps and attempt alternative discovery methods.

IF VALIDATION FAILS:

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

IF STILL INCOMPLETE after 2 iterations:

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /metrics — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.