AI Workflow Templates
QualitystarterCI-safeupdated 2026-08-11

Prompt Review Gate

Run ARC Prompt Reviewer as a CI gate: score clarity, ambiguity, injection surface, and failure modes before a prompt ships.

#prompts#ci#quality#sister-tool
View source →

Tools in the pipeline

  • validator
    prompt-reviewer
    Static review. No model call.
  • api
    ci
    Fail the job when grade is below --min-grade.
Stack: Prompt Reviewer CLI · GitHub Actions

What it does

Wraps the free ARC Prompt Reviewer as a workflow step. Paste or path a system prompt, get a 0-100 score, letter grade, and concrete rewrites. Fail CI when the grade is below the bar. Static and reproducible. No model call.

Pipeline

  1. Collect — prompt file (and optional tool list) in the PR.
  2. Reviewnpm run review -- path/to/prompt.txt --min-grade=B.
  3. Report — Markdown/JSON on the PR.
  4. Gate — fail the check under the minimum grade.
  5. Rewrite — apply suggested blocks and re-run.

Prompt skeleton

npm run review -- prompts/system.md --tools tools.txt --min-grade=B --json

Guardrails

A high grade means the prompt states the right defenses, not that a live model is unbreakable. Always red-team before production. Sister tool: github.com/aking-beep/prompt-reviewer.

Structural dry-run

No model call. Validates required fields only.

Prefills the example payload for Support copilot prompt. Edit it, then run.

Related