---

name: track-metrics description: Record task metrics after completing work. Use after /verify or any completed task to build measurement data.

Track Task Metrics

Record objective task outcomes to measure if the repo's systems actually improve results.

When to Invoke

• After /verify completes
• After any task completes (success or failure)
• After debugging sessions
• When hooks block something

Instructions

Step 1: Gather Task Data

Collect the following (ask user if unclear):

Outcome Data:

Process Quality Data:

Escalation Criteria:

• appropriate: Used debug-fresh or asked for help at attempt 3
• late: Went 5+ attempts before escalating
• never: Should have escalated but didn't
• unnecessary: Escalated on attempt 1 (over-cautious)
• na: Task didn't require escalation consideration

Step 2: Read Current Metrics

CopyRead: .claude/metrics/metrics.json

Step 3: Create Task Entry

Format the new task:

Copy{
  "id": "[next sequential ID]",
  "date": "[ISO date: YYYY-MM-DD]",
  "task_type": "[type]",
  "outcome": "[success|partial|failure]",
  "description": "[one-line summary]",
  "skills_used": ["skill-1", "skill-2"],
  "hooks_triggered": {
    "blocks": ["hook-name"],
    "warnings": ["hook-name"]
  },
  "verification": {
    "tests": "[pass|fail|skipped]",
    "lint": "[pass|fail|skipped]"
  },
  "attempts": [number],
  "process": {
    "first_hypothesis_correct": [true|false],
    "plan_deviations": [number],
    "user_corrections": [number],
    "escalation": "[appropriate|late|never|unnecessary|na]",
    "skill_adherence": "[followed|partial|skipped|na]"
  }
}

Step 4: Update Metrics File

Add the task to the tasks array and update the summary:

1. Append task to tasks array
2. Increment summary.total_tasks
3. Recalculate summary.success_rate = (successes / total) * 100
4. Update summary.by_type counts
5. Update summary.by_skill counts
6. Update summary.hooks_triggered totals
7. Recalculate summary.process_quality scores:
   • first_try_correct_rate = (tasks with first_hypothesis_correct=true) / total * 100
   • avg_plan_deviations = sum of plan_deviations / total
   • avg_user_corrections = sum of user_corrections / total
   • escalation_appropriate_rate = (appropriate + na) / total * 100

Use the Edit tool to update .claude/metrics/metrics.json.

Step 5: Report

Output a brief confirmation:

CopyMetrics Recorded:
- Task: [description]
- Outcome: [outcome]
- Type: [task_type]
- Attempts: [N]
- Process: hypothesis=[correct/wrong], deviations=[N], corrections=[N]

Running Totals:
- Total tasks: [N]
- Success rate: [X]%
- Process quality: [X]% (first-try correct rate)
- Most used skill: [skill-name] ([N] uses)
- Hooks blocked: [N] times

Objective Criteria

Success = Task completed, verification passed, no rework needed Partial = Task completed but with caveats (skipped tests, known issues) Failure = Task not completed, blocked, or abandoned

Attempts:

• 1 = First implementation worked
• 2+ = Required iteration/debugging
• Count major rework cycles, not minor edits

What NOT to Track

• Subjective quality assessments
• Time spent (unreliable)
• "How good" the code is
• Feelings about the task

Only track what's objectively observable.

Example

Copy{
  "id": "001",
  "date": "2026-01-06",
  "task_type": "feature",
  "outcome": "success",
  "description": "Added MCP security validation hook",
  "skills_used": ["skill-creator"],
  "hooks_triggered": {
    "blocks": [],
    "warnings": []
  },
  "verification": {
    "tests": "skipped",
    "lint": "pass"
  },
  "attempts": 1,
  "process": {
    "first_hypothesis_correct": true,
    "plan_deviations": 0,
    "user_corrections": 0,
    "escalation": "na",
    "skill_adherence": "followed"
  }
}

Why This Matters

Without metrics, "improvement" is just vibes. With metrics:

• We know if success rate is trending up or down
• We know which skills correlate with success
• We know if hooks are catching real issues
• We can A/B test changes to skills/rules