Home/Quality

Why these briefs don’t rot

Prompt catalogs age badly: rules go stale, quality varies by entry, and nothing enforces a floor. Every one of the 157 briefs here passes a published linter and a CI gate before it ships — this page is the bar, with links to the code that enforces it.

Browse the catalog Read the linter source ↗

The linter

Every brief passes the same published linter.

Not a style guide — a build step. python3 build.py refuses to build the site (and the deploy) unless every brief in the catalog clears every rule below. The rules live in build.py, in the open, and the linter has its own test suite so the bar itself can’t silently loosen.

Structure — enforced on all 157 Goal Prompts

The four-phase skeleton: ## Phase 1–4 — Explore → Audit → Curate → Report — plus a ## Rules section.
Body under 4,000 characters — short enough to read before you run it.
Phase 2 checks through 4–12 named lenses; each report section has a defined shape.
Each Goal Prompt names its one report file (REPORT.md-shaped, ALL-CAPS), and no two may write the same file.
Re-runs are first-class: the report is dated, and a report that already exists is read first so the new one leads with what changed.

Safety — the ask-first gate

Every Goal Prompt ends with the literal gate: “Report only — end by asking …” — the linter greps for it and fails the build without it.
Goal Prompts are read-only toward your code; the only write is the report itself (honest exception: 47 · The Fixer modifies code, and gates on your explicit selection first).
A Goal Prompt whose subject a repo simply doesn’t have must say so in a one-paragraph null report and stop — no invented findings.

The gate

CI runs the whole bar on every change.

Every push and pull request runs .github/workflows/ci.yml: the build (which lints all 157 briefs), the linter’s own tests, a hermetic installer test, and the MCP smoke test — then fails on any drift between sources and generated files. The Vercel deploy runs the same build as a hard gate, so an oversized or rule-breaking brief can’t reach production. Contributions clear the identical bar — the linter is the moderation story (CONTRIBUTING.md).

The receipts

It’s run against its own code, in the open.

The briefs audit this site’s own repo: the reports they wrote are published as live samples, and the FIXLOG traces shipped fixes back to the report and finding that surfaced each one. Nothing here is a mock-up — the loop (brief → report → Studio → Fixer → FIXLOG) is how this catalog maintains itself. When a brief under-delivers on this repo, it gets sharpened before it stays in the catalog.