Stats Tracker

Skill for tracking Claude Code usage statistics AND skill effectiveness (RSI metrics).

When to Use

This skill activates when you:

Want to see your usage statistics
Track productivity over time
Analyze model usage patterns
Monitor usage streaks
Plan based on usage limits
NEW: Review skill effectiveness metrics
NEW: Generate weekly effectiveness reports

Keywords: stats, usage, analytics, productivity, streak, model usage, tokens, sessions, cost, skill effectiveness, skill metrics

Quick Commands

Command	Description
`/stats`	View usage stats, streak, favorite model
`/stats skills`	View skill effectiveness dashboard
`/stats skills [name]`	Deep-dive on specific skill
`/stats verification`	NEW: View pass@k metrics by verification type
`/stats verification log`	NEW: Log a verification session result
`/usage`	View token usage and limits
`/context`	View current context usage
`/cost`	View session cost breakdown

Understanding /stats Output

/stats

📊 Your Claude Code Stats
────────────────────────────
🏆 Favorite Model: Sonnet 4 (78% of sessions)
🔥 Current Streak: 12 days
📈 This Week: 847K tokens across 23 sessions
📉 Usage Graph: [▁▃▅▇█▆▄▂] (last 7 days)

Metrics Explained

Metric	Description
Favorite Model	Most frequently used model
Current Streak	Consecutive days using Claude Code
This Week	Total tokens and sessions
Usage Graph	Visual of daily usage pattern

Understanding /usage Output

/usage

📊 Usage This Period
────────────────────────────
Sonnet: ████████░░ 78%
Opus:   ██░░░░░░░░ 15%
Haiku:  █░░░░░░░░░ 7%

Resets: Monday 00:00 UTC

Model Selection Guide

Choose the right model for the task:

Task Type	Model	Why
General coding	Sonnet	Best balance of speed/quality
Complex architecture	Opus	Deep reasoning, long context
Quick fixes	Haiku	Fast, efficient
Code exploration	Haiku (Explore)	Optimized for search
Documentation	Sonnet	Good writing quality
Bug investigation	Sonnet/Opus	Depends on complexity

Cost Efficiency

Model	Relative Cost	Best For
Haiku	$	Quick tasks, exploration
Sonnet	$$	Most development work
Opus	$$$	Complex problems

Daily Workflow with Stats

Morning Routine

1. /stats                 # Check streak, plan day
2. /usage                 # Check remaining budget
3. /rename feature-x      # Name today's session
4. Start coding           # Use appropriate model

During Development

# After complex task
/context                  # Check context usage

# When switching tasks
/stats                    # Quick progress check

End of Day

/stats                    # Review productivity
/usage                    # Check usage remaining
/cost                     # See session cost

Usage Optimization Strategies

If Usage is High

Use Haiku more - For exploration and quick tasks
Spawn Explore agents - Uses efficient Haiku model
Background tasks - Don't consume extra tokens
Be more targeted - Specific queries use less context

If Starting Fresh Week

Plan major tasks - Allocate Opus for complex work
Default to Sonnet - Best general-purpose model
Reserve Haiku - For quick lookups and exploration

Maximizing Efficiency

# Good: Specific query
"Fix the null check on line 45 of auth.ts"

# Bad: Vague query (uses more context/tokens)
"Help me with the auth system"

Tracking by Feature

Use named sessions to track usage per feature:

# Start feature work
/rename billing-feature

# Work on feature...

# Check stats
/stats  # See usage for this session

# Compare features
"How many tokens did billing vs auth use?"

Weekly Review Pattern

Every Friday:

1. /stats              # Review week's usage
2. /usage              # Check budget status
3. Note completions    # What features finished?
4. Plan next week      # Allocate model usage

Weekly Goals Template

Week Goals:
- [ ] Maintain 5-day streak
- [ ] Complete 3 features
- [ ] Use Opus only for architecture decisions
- [ ] Try Explore agent for research tasks
- [ ] Stay under 80% usage

Integration with Other Skills

With Session Manager

# Track stats per named session
/rename billing-feature
[work on feature]
/stats
# See usage specific to this session

With Context Manager

# Monitor both context and usage
/context  # Current context budget
/stats    # Overall usage patterns

With Async Runner

# Background tasks don't inflate usage
"Run build in background"
# Build output doesn't use tokens

Gamification Ideas

Streak Challenges

5-day streak: Bronze
10-day streak: Silver
30-day streak: Gold
Try to never break your streak!

Model Efficiency Challenge

Challenge: Complete task with Haiku
1. Try Haiku first for any task
2. Escalate to Sonnet only if needed
3. Use Opus sparingly for max impact

Weekly Competition (Team)

Compare completed features
Compare usage efficiency
Share tips for optimization

External Tools

ccusage (CLI Analytics)

# Install
npm install -g ccusage

# View usage
ccusage

# Daily breakdown
ccusage --daily

# Monthly view
ccusage --monthly

Claude Code Usage Monitor

Real-time terminal monitoring with:

Token tracking
Burn rate analysis
Predictions for limits

CircleTel Productivity Patterns

Feature Development Session

Start: /rename feature-customer-billing
       /stats  # Baseline

Middle: [coding work]
        /context  # Check context

End: /stats  # Compare to baseline
     /cost   # Session cost

Bug Fix Session

/rename BUG-1234-fix
/stats  # Note starting point

[investigation and fix]

/stats  # See investigation cost
# Typically: Sonnet for analysis, quick fixes

Architecture Planning

/rename architecture-review

# Use Opus for deep thinking
"Think about the best approach for..."

/stats  # Higher usage expected
/cost   # Worth it for good architecture

Skill Effectiveness Tracking (RSI Metrics)

NEW in v2.0: Track which skills perform best and feed insights back for improvement.

Command: `/stats skills`

/stats skills

═══════════════════════════════════════════════════════════════
  SKILL EFFECTIVENESS DASHBOARD - February 2026
═══════════════════════════════════════════════════════════════

MOST EFFECTIVE (by success rate)      │ NEEDS IMPROVEMENT
──────────────────────────────────────┼────────────────────────────
1. database-migration      95% ✓      │ 1. coverage-check      70% ⚠
2. bug-fixing              89% ✓      │ 2. refactor            75% ⚠
3. compound-learnings      87% ✓      │
                                      │
MOST USED (activations)               │ UNDERUTILIZED
──────────────────────────────────────┼────────────────────────────
1. bug-fixing              47 times   │ 1. mobile-testing       2 times
2. context-manager         35 times   │ 2. deployment-check     3 times
3. stats-tracker           28 times   │

INSIGHT: bug-fixing success rate improved 12% after adding
        error-registry integration last week.

RECOMMENDATION: Consider promoting mobile-testing skill
               (high success rate, low activation).
═══════════════════════════════════════════════════════════════

Metrics Tracked Per Skill

Metric	Description	How Measured
Activation Count	Times skill triggered	Keyword detection
Success Rate	Completed without correction	No follow-up corrections
Resolution Time	From activation to completion	Session timestamps
Correction Rate	How often skill was corrected	Links to compound-learnings
Pattern Contribution	New patterns generated	Links to learnings/

Deep-Dive: `/stats skills [name]`

/stats skills bug-fixing

═══════════════════════════════════════════════════════════════
  SKILL DEEP-DIVE: bug-fixing
═══════════════════════════════════════════════════════════════
Version: 1.1.0
Dependencies: error-registry

METRICS (February 2026)
────────────────────────────────────────────────────────────────
Activations:        47
Success Rate:       89%
Avg Resolution:     23 min
Corrections:        7 (15%)
Patterns Created:   3

PHASE BREAKDOWN
────────────────────────────────────────────────────────────────
Phase 0 (Registry):   Avg 2min  │ Skip rate: 20%
Phase 1 (Understand): Avg 5min  │ Skip rate: 10%
Phase 2 (Investigate): Avg 10min │ Skip rate: 5%
Phase 3 (Fix):        Avg 5min  │ Success: 95%
Phase 4 (Validate):   Avg 3min  │ Skip rate: 30%

TREND
────────────────────────────────────────────────────────────────
Success rate: ↑ 12% since error-registry integration
Registry hits: 40% of bugs found in known patterns

RECENT CORRECTIONS (to improve)
────────────────────────────────────────────────────────────────
- 2026-02-10: Missed RLS policy check
- 2026-02-08: Used wrong column name
═══════════════════════════════════════════════════════════════

Data Storage

Skill metrics are stored in:

.claude/skills/stats-tracker/
├── metrics.json              # Current period metrics
├── verification-metrics.json # Pass@k verification metrics
└── reports/
    └── weekly-YYYY-WW.md     # Weekly reports

Verification Metrics (Pass@k Tracking)

Track how many verification attempts are needed before success.

Command: `/stats verification`

/stats verification

═══════════════════════════════════════════════════════════════
  VERIFICATION METRICS - March 2026
═══════════════════════════════════════════════════════════════

FIRST-ATTEMPT SUCCESS (pass@1)
────────────────────────────────────────────────────────────────
Type-check:  ████████░░ 84% (target: 90%)
Build:       ███████░░░ 78% (target: 85%)
Unit tests:  ██████░░░░ 72% (target: 75%) ✓
E2E:         █████░░░░░ 58% (target: 60%)
Lint:        █████████░ 96% (target: 95%) ✓

TREND (pass@1 rate, last 4 weeks)
────────────────────────────────────────────────────────────────
Week 10: ▃ 78%
Week 11: ▅ 82%
Week 12: ▇ 85% ↑

COMMON FIX CATEGORIES
────────────────────────────────────────────────────────────────
1. Code changes:       62%
2. Config tweaks:      18%
3. Missing imports:    12%
4. Test adjustments:   8%

INSIGHT: Type-check pass@1 improved 7% after adding
        type-guards-optionals.md rule.
═══════════════════════════════════════════════════════════════

Understanding Pass@k

Metric	Meaning	Target
pass@1	Passed on first attempt	> 80%
pass@3	Passed within 3 attempts	> 95%
pass@5	Passed within 5 attempts	100%
fail@5+	Needed 5+ attempts	< 2%

Grader-Specific Targets

Verification Type	pass@1 Target	Rationale
type-check	90%	Type errors are predictable
build	85%	Env/import issues common
unit-test	75%	Logic bugs take iteration
e2e	60%	Integration complexity
lint	95%	Mostly auto-fixable
manual	80%	Subjective criteria

Logging a Verification Session

After completing verification, log the result:

/stats verification log type-check 2 code

# Logs: type-check verification, passed on attempt 2, fix was code-related

Format: /stats verification log <type> <attempts> <fix_category>

Fix Categories

Category	Examples
`code`	Wrong type, missing property, logic error
`config`	tsconfig, package.json, env vars
`dependency`	Missing import, version mismatch
`test`	Test assertion wrong, fixture issue

RSI Integration

Low pass@1 rates trigger investigation:

pass@1 < target
      │
      ▼
Analyze fix_categories
      │
      ▼
Common pattern? ───► Create compound-learning
      │
      ▼
Extract rule ────────► Add to .claude/rules/
      │
      ▼
pass@1 rate improves

Weekly Verification Review

Add to your Friday review:

/stats verification — Check pass@1 rates
Compare to targets — Which types need improvement?
Review fix categories — What's causing failures?
Consider new rules — Can a pattern prevent failures?

Weekly Report Generation

Every Monday, generate a skill effectiveness report:

# Skill Effectiveness Report - Week 7, 2026

## Summary
- Total skill activations: 142
- Average success rate: 85%
- Most improved: bug-fixing (+12%)
- Needs attention: coverage-check (70%)

## RSI Loop Impact
- Error registry hits: 40% (saving ~10min per known bug)
- Corrections captured: 5
- Rules extracted: 2
- Patterns created: 3

## Recommendations
1. Add more MTN API patterns to error-registry
2. Update coverage-check with fallback providers
3. Promote mobile-testing (underutilized, high success)

Integration with RSI Skills

┌─────────────────────────────────────────────────────────────────┐
│                    RSI METRICS FLOW                              │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  SKILL ACTIVATION ──► TRACK METRICS ──► WEEKLY ANALYSIS         │
│         │                                      │                 │
│         │                                      ▼                 │
│         │                            GENERATE INSIGHTS           │
│         │                                      │                 │
│         │              ┌───────────────────────┴───────┐         │
│         │              │                               │         │
│         ▼              ▼                               ▼         │
│  CORRECTION? ──► compound-learnings            SKILL UPDATE      │
│                        │                               │         │
│                        ▼                               │         │
│                 error-registry ◄───────────────────────┘         │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Best Practices

Check stats daily - Start with /stats
Match model to task - Don't use Opus for simple fixes
Track by feature - Use named sessions
Review weekly - Adjust workflow based on patterns
Use Haiku for exploration - Spawn Explore agents
Background for builds - Don't waste tokens on output
Targeted queries - Specific questions = efficient usage
NEW: Review /stats skills weekly for improvement opportunities

Troubleshooting

Stats seem wrong

# Check current session specifically
/cost

# Check usage for period
/usage

High usage unexpectedly

Check for large file reads
Review if background tasks are actually background
Consider if queries could be more specific

Streak reset

Streak requires at least one interaction per day
Even /stats counts
Check timezone (UTC-based)

Version: 2.0.0 Last Updated: 2026-02-12 For: Claude Code v2.0.64+ RSI Integration: error-registry, compound-learnings

stats-tracker