prompt-audit
Skill10 starsChanged 4 months agoMITView on GitHub
Audit a prompt library as an operating system, not a pile of prompts. Use to find what a prompt collection over-does, under-does, and where it leaks into meta-looping. Classifies each prompt by the work it performs, finds overrepresented and missing moves, flags prompts that cause meta-loops or never produce artifacts, and returns the minimum upgraded set as a sequenced workflow.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- fedecaccia/prompt-audittheir own version, first seen Nov 8, 2026
- ejoliet/claude-skillstheir own version, first seen Oct 11, 2026
- PatriotAgenticLLC/better_claude_skillztheir own version, first seen Jul 1, 2026
- Sharan0516/claude-token-optimizertheir own version, first seen Apr 13, 2026
- JWhiteUX/ClaudeSkills-Librarytheir own version, first seen Apr 2, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- JWhiteUX/ClaudeSkills-Library★ 4same name
- PatriotAgenticLLC/better_claude_skillz★ 3same name
- fedecaccia/prompt-audit★ 2same name
- ejoliet/claude-skills★ 0same name
- Sharan0516/claude-token-optimizer★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository24 other files it gives its agents.
The file
Files
/ratchet:prompt-audit — the library as an OS
A prompt library is not a pile of good words; it is an operating system for cognition. This command audits it as one. It does not rewrite prompts first — it diagnoses the system: what work each prompt does, which moves are missing, and where the library lets the model escape into elegant analysis instead of producing something.
Step 0 — Load the library
Take the prompt list from the argument, a file, or the current context. Do not rewrite yet.
Classify
Tag each prompt by the type of work it performs (a prompt may do more than one):
- target locking · diagnosis · artifact production · adversarial testing · decision · patching · memory/state · orchestration · compression · boundary pushing
Find the system's shape
- Overrepresented moves — where the library is thick (usually diagnosis and "next move" prompts).
- Missing moves — where it is thin (usually artifact production, patching, testing, serialized state).
- Meta-looping prompts — those that ask what to ask next without ever forcing a move. These are the hallway of mirrors.
- Artifact-producing prompts — the ones that actually leave something behind.
- Prompts to delete or merge — redundant, vague, or low-consequence.
- The minimum new prompt set that would outperform the current library.
The operating rule
Judge every prompt against this bar. Keep it only if it does at least one:
- Forces a choice. 2. Creates an artifact. 3. Tests an artifact. 4. Patches a defect.
- Serializes useful state. 6. Kills a bad option. 7. Pushes to a higher-yield move.
Everything else is smoke.
Output contract
CLASSIFICATION: <table: prompt → work type(s)>
OVERREPRESENTED: <moves the library over-does>
MISSING: <moves it lacks — the expensive gaps>
META-LOOPERS: <prompts that never force a move>
DELETE / MERGE: <the cuts>
UPGRADED SET (sequenced workflow, not a pile):
1. <prompt> — <the move it forces>
2. ...Serialize
ratchet artifact add '{"title":"prompt-library audit","kind":"decision-record","status":"v0","holes":["..."]}'Produce the upgraded library as a sequenced workflow — frame → choose → build → attack
→ patch → serialize → advance — never a flat list. Next: /ratchet:build the new set.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

