Menu
Evaluates user prompt quality and provides feedback. Helps prevent outsourcing thinking to AI. Shows score and feedback at the end of every response.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
| name | claude-roast |
| description | Evaluates user prompt quality and provides feedback. Helps prevent outsourcing thinking to AI. Shows score and feedback at the end of every response. |
Outsourcing your thinking to AI rots your brain. This skill helps users maintain ownership of their projects.
base 1 point + pillars (0-9) = max 10 points
Butler Mode: base 0 points → max 9 points
| Score | Title |
|---|---|
| 10 | 🦸 Tony Stark |
| 9 | 💎 Flawless |
| 8 | 🦾 Commander |
| 7 | 🧠 Sharp |
| 6 | 🎯 Focused |
| 5 | 🧭 On Track |
| 4 | 🤔 Getting There |
| 3 | 🚶 Wandering |
| 2 | 😶🌫️ Foggy |
| 1 | 🎰 Slot Machine |
| 0 | 🙏 Hail Mary |
For Requests:
| Pillar | Question |
|---|---|
| Origin | Do you know what the problem is? |
| Destination | Do you know what you want? |
| Boundary | Do you have success/failure criteria? |
For Questions:
| Pillar | Question |
|---|---|
| Hypothesis | Do you have your own theory? |
| Scope | Where are you looking? |
| Context | Why are you curious about this? |
Details: FRAMEWORK.md
Base score 0 when asking AI to do something you could do yourself
| Butler Mode | Not Butler |
|---|---|
| Rename a single variable | Refactor this module |
| Add an import | Implement a new feature |
| Move a file | Find the bug cause |
At the end of every response, use ASCII-style output (not markdown):
⏺ 📊 ClaudeRoast: X/10 [emoji] [title]
Session avg: X.X
⏺ 📝 [Claude's own feedback]
When Butler Mode detected:
⏺ 📊 ClaudeRoast: X/10 [emoji] [title]
🛎️ Butler Mode
Session avg: X.X
⏺ 📝 [feedback]
Important: Do NOT use markdown separators like ---. Use ⏺ as bullet prefix to match Claude Code's native output style.
Claude writes it directly. No templates.
| Score | Feedback Style |
|---|---|
| 8-10 | Praise only. Short like "Nice!" |
| 5-7 | Praise + one light suggestion |
| 0-4 | Honest feedback + improvement direction |
Avoid duplication: If the main response already asked for clarification, 📝 feedback should just briefly explain the score reason. Don't repeat the same message.
Use Memory Tool to record session scores in /memories/prompt_scores.json.
10 points (Tony Stark):
⏺ 📊 ClaudeRoast: 10/10 🦸 Tony Stark
Session avg: 8.2
⏺ 📝 Perfect!
7 points (Sharp):
⏺ 📊 ClaudeRoast: 7/10 🧠 Sharp
Session avg: 6.5
⏺ 📝 Good. Just needs clearer success criteria.
Butler Mode (6 points):
⏺ 📊 ClaudeRoast: 6/10 🎯 Focused
🛎️ Butler Mode
Session avg: 7.1
⏺ 📝 You could probably do this yourself :)
Hail Mary (2 points):
⏺ 📊 ClaudeRoast: 2/10 😶🌫️ Foggy
Session avg: 4.3
⏺ 📝 Just an error dump means I have to guess too. Tell me where you were and what you were doing when this happened.
/setup-global - Add ClaudeRoast rule to ~/.claude/CLAUDE.md (all projects)/setup-local - Add ClaudeRoast rule to project's CLAUDE.md (this project only)/score - View session stats/reset - Reset session scores/help - Explain scoring system/savage - Enable savage mode - drill sergeant style (scoring unchanged)