← ClaudeAtlas

auditlisted

Run /audit to write the AGENTS.md context files every other skill reads: the real stack, the commands that actually work, the folder layout, the conventions the code actually follows. Run it first on an existing codebase, or after the stack is scaffolded on a new one. Records only what it can see evidence for, and never overwrites anything a human wrote.
rishabhsisodiya/stop-guessing · ★ 0 · Code & Development · score 70
Install: claude install-skill rishabhsisodiya/stop-guessing
## What this skill does Writes down what is true about this project, so every other skill stops guessing. `/develop` reads `AGENTS.md` to know the test command. `/architect` reads it to know the patterns in use. `/verify` reads it to know how to start the app. Without it, each of them re-derives your project from scratch every session, and gets it slightly wrong each time. The output is `AGENTS.md` at the repo root, plus nested ones where a subdirectory genuinely differs, plus a thin `CLAUDE.md` that points at it so Claude Code picks it up too. ## Facts, not standards This is the line that keeps `AGENTS.md` useful: - **A standard** is how code should be written, is true of every project, and already ships in `develop/build-standards.md`. Do not copy any of it here. - **A fact** is what is true of *this* repo, and can only be discovered by reading it. `AGENTS.md` holds facts. "No query inside a loop" is a standard and does not belong here. "Tests are `pnpm test:unit`, and services live in `src/modules/<name>/`" is a fact and does. If you cannot point at the evidence, it is not a fact. ## What this skill refuses to do 1. **Never records a convention without evidence.** Two or more real examples, or it does not go in. One occurrence is a sighting, not a convention. 2. **Never writes an aspiration.** If the repo has three tests and no CI, `AGENTS.md` says that, not "we practise TDD". A context file that describes a better project than the one on disk makes ev