Wikiprompt
Talk

Codex Skill File Audit Prompt

From Wikiprompt, the free prompt encyclopedia

Alvaro Cintas
Contributed byAlvaro CintasXSource

Sep 7, 2026

Codex Skill File Audit Prompt A prompt for auditing skill files and AGENTS.md against six best-practice criteria, flagging outdated instructions and providing structured output.

Prompt ContentSave

🌐
Codex, go through every skill file and every AGENTS.md file inside this project and check each one against these six things, based on what OpenAI's codex dx found when reviewing best practices for GPT-6 Astra. Don't rewrite anything yet, just report back. 1. BLOATED OR VAGUE SKILL DESCRIPTIONS Flag any skill description that's too long or too broad. Long descriptions risk getting truncated when too many skills are loaded, and vague ones can get picked for tasks they don't actually fit. 2. MISSING PROGRESSIVE DISCLOSURE Flag any single long skill file covering multiple workflows instead of a short router pointing to separate supporting docs, only loaded when needed. 3. RIGID STEP-BY-STEP RECIPES Flag any skill written as an elaborate, overly specific itinerary, do this then this then this, instead of trusting the model's own judgment on how to handle it. 4. FORCING FULL CONTEXT BEFORE EVERY EDIT Flag any AGENTS.md instruction requiring a full repo read or a stack of docs before every edit, even a small one. 5. OUTDATED TESTING OR VERIFICATION PUSHES Flag any instruction pushing the model to run tests or double-check itself in ways it would already do automatically now. 6. OVER-CAUTIOUS PERMISSION LANGUAGE Flag any strong "ask before doing X" language that might now be making the model stop too early or ask for approval on something it should safely continue. For each thing you flag, give me: FLAGGED - [which of the 6] - [the instruction, and which file it's in] Why: [what old problem this was fixing] What to do: [delete it, shorten it, split it into a router, or rewrite it] For anything that should stay: KEEP - [the instruction] Why: [what real, lasting thing this describes] Don't flag something just because it's specific, if it's a real, permanent fact about this project, a security boundary, a team standard, it stays. Only flag what exists purely to compensate for an older model. End with one line: roughly what percent of what you reviewed falls into each of the 6 flagged types versus how much should stay

Sign in to see the full prompt

Continue with:

By logging in, you agree to our Terms of Use and Privacy Policy

Usage

This prompt is designed for use with productivity. Copy the prompt content above and paste it into your preferred AI tool.

For best results, you may customize the placeholders (indicated by square brackets or capital letters) with your specific requirements.

References

Categories:productivity| twitter| codex| skill-audit

Talk