Skip to content
ai-contributors logomark ai-contributor-spec / audit
Audit › No-install prompt

AI Contributor Audit Prompt

For repeat audits, prefer the ai-contributor-audit skill. The skill uses the same audit model, but is easier to reuse and less error-prone.

Copy-Paste Prompt

Section titled “Copy-Paste Prompt”

Run this from the repository root:

The audit prompt14 linesDownload .txt
Audit this repository against the AI Contributor Specification.


1. Resolve the current specification source to a full commit SHA:
   git ls-remote https://github.com/ai-contributors/ai-contributor-spec.git refs/heads/main
   Capture the 40-char SHA as SPEC_SOURCE_COMMIT.


2. Download the pinned runbook with the bootstrap script. This file list is authoritative; do not guess paths from prose:
   curl -fsSL \
     "https://raw.githubusercontent.com/ai-contributors/ai-contributor-spec/${SPEC_SOURCE_COMMIT}/skills/ai-contributor-audit/scripts/bootstrap.ts" \
     -o /tmp/ai-contributor-bootstrap.ts
   npx --yes tsx@4.21.0 /tmp/ai-contributor-bootstrap.ts "${SPEC_SOURCE_COMMIT}"
   The bootstrap writes the pinned runbook outside the repository, normally under /tmp, and prints the path of the SKILL.md to follow next.


3. Follow that SKILL.md exactly. Do not guess file paths or fetch sources from a different ref. Base every status on the current working directory, generated evidence, or recorded repository settings -- never on memory. Do not update the audit artifacts as standalone Markdown files; status, summary, and backlog changes must go through collect, stamp, evidence review for judgment-required rows, stamp, and validate. Treat the filled checklist and audit log as reviewable audit output, not human/accountable-owner acceptance.

Review Before Trust

Section titled “Review Before Trust”

For agent-run audits, treat the filled checklist and audit log as reviewable artifacts. Agent audit results can be wrong or invented, even when a runbook and evidence collector are used.

Before you commit filled audit files:

  • Check a sample of Fulfilled rows against the cited evidence.
  • Review every Not relevant row and confirm the reason applies.
  • Re-run any standard command listed by the audit runbook.

The audited_commit and auditor fields exist so a reviewer can reproduce and challenge any line.

For the evidence model behind this flow, read AI-CONTRIBUTOR-AUDIT-MODEL.md.