skill-compass
VerifiedEvaluate skill quality, find the weakest dimension, and apply directed improvements. Also tracks usage to spot idle or risky skills. Use when: first session after install, or user asks about skill quality, evaluation, inbox, suggestions, or improvement.
examples/guide-claudeception.md
View on GitHub# SkillCompass + Claudeception
> **Quality control for skill extraction.**
> Claudeception turns good debugging sessions into reusable skills. SkillCompass makes sure they're actually good.
---
## The Problem
Claudeception is powerful — it watches your debugging sessions and automatically extracts reusable skills. But extraction quality varies wildly:
- **Session-specific artifacts**: hardcoded file paths, user-specific config references
- **Security holes**: API keys from the session, unrestricted shell access patterns
- **Redundancy**: extracted skill duplicates one you already have (or a public skill on ClawHub)
- **Incomplete logic**: the debugging session solved one case, but the extracted skill doesn't generalize
Without quality control, your `skills/` directory grows into a pile of untested drafts. Some are gems. Some are liabilities. **You can't tell which is which.**
## The Solution
Install both. SkillCompass automatically evaluates every new skill Claudeception creates:
```
You debug a tricky deployment issue in a session
→ Claudeception extracts the pattern
→ Writes skills/deploy-fixer/SKILL.md
SkillCompass detects new SKILL.md (first-creation trigger)
→ Automatic six-dimension evaluation:
╭──────────────────────────────────────────────╮
│ SkillCompass — Skill Quality Report │
│ deploy-fixer · v1.0.0 · atom │
├──────────────────────────────────────────────┤
│ D1 Structure ████████░░ 8/10 │
│ D2 Trigger █████░░░░░ 5/10 │
│ D3 Security ██░░░░░░░░ 2/10 ⛔ CRIT │
│ D4 Functional ██████░░░░ 6/10 │
│ D5 With/Without +0.18 │
│ D6 Uniqueness ████░░░░░░ 4/10 │
├──────────────────────────────────────────────┤
│ Overall: 44/100 · Verdict: FAIL │
│ Weakest: D3 Security — hardcoded path to │
│ /Users/dev/.ssh/config │
│ D6 flag: 85% overlap with shell-helper │
╰──────────────────────────────────────────────╯
```
## What Happens Next
The evaluation tells you exactly what to fix and in what order:
```bash
# Fix the security issue first (D3 is gate dimension)
/eval-improve ./skills/deploy-fixer/SKILL.md
→ Removes hardcoded path, adds user confirmation step
→ D3: 2 → 7 ✓
→ v1.0.0-evo.1 saved
# Address the redundancy (D6)
/eval-compare ./skills/deploy-fixer/SKILL.md ./skills/shell-helper/SKILL.md
→ Side-by-side comparison shows 85% overlap
→ Unique value: deploy-fixer handles pm2 restart + health check
→ Recommendation: merge deploy-specific logic into shell-helper,
or narrow deploy-fixer scope to deployment-only tasks
```
## The Extraction → Quality Pipeline
```
Claudeception extracts skill
↓
Pre-Accept Gate (immediate)
→ D1 + D3 lightweight check
→ Critical issues flagged instantly
↓
First-creation evaluation (automatic)
→ Full six-dimension assessment
→ Redundancy check against existing skills
↓
/eval-improve (your decision)
→ Fix weakest dimension
→ Verify improvement
→ Snapshot saved
↓
Production-ready skill
```
## Common Patterns
| Claudeception extracts... | SkillCompass catches... | Fix |
|--------------------------|----------------------|-----|
| Skill with hardcoded paths | D3: hardcoded file system paths | `/eval-improve` parameterizes paths |
| Skill that duplicates existing | D6: high overlap with installed skill | Merge or discard |
| Skill with vague description | D2: low trigger accuracy | `/eval-improve` sharpens description |
| Skill that only handles one case | D4: poor edge case coverage | `/eval-improve` generalizes logic |
| Actually good skill | All dimensions green | Silent pass, snapshot saved |
## Setup
```bash
# 1. Install SkillCompass
clawhub install skill-compass
# or: cp -r skill-compass/ ~/.claude/skills/skill-compass/
# 2. Install Claudeception (if not already)
# Follow Claudeception's own install instructions
# 3. Done. SkillCompass watches the skills/ directory.
# Every new SKILL.md triggers automatic evaluation.
```
No integration config. Claudeception writes files, SkillCompass watches files. They don't need to know about each other.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 470k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |