Spec Quality Audit: AI Skill for Product Management
Spec Quality Audit is an AI skill that audits specs and PRDs for untestable requirements, missing sections, and happy-path-only flows. Every finding carries a severity rating and a concrete rewrite.
npx skills add Uxcel-Lab/product-skills --skill pm-spec-quality-audit
What this skill does
Audits specs, PRDs, and requirements docs for the gaps that make teams build the wrong thing.
- Checks the spec opens with the problem and every requirement traces to a goal.
- Keeps each document at its own layer: requirements, functional behavior, or technical detail.
- Tests every requirement with one question: could QA write a pass/fail check from it?
- Demands complete flows, including errors, interruptions, and returning users.
- Checks scope states what is out, not just what is in.
- Verifies dependencies have owners and assumptions read as testable statements.
- Asks for evidence behind every "validated with users" claim.
- Rates findings by build damage and ends with the top three rewrites.
When to use it
Use it when you need an AI assistant to:
- audit a PRD or spec before development starts;
- catch untestable requirements like "fast" and "intuitive";
- find what the document is missing, not just what is weak;
- check error flows and corner cases got specified;
- state scope exclusions before misalignment hides in them;
- validate a spec generated by an AI assistant.
What this audit checks
| Category | What it catches |
|---|---|
| Problem anchoring | Specs that open with the solution, requirements serving no stated goal |
| Layer discipline | Architecture leaking into functional specs, outcome wishes with no behavior |
| Testability | "Fast" and "intuitive" with no number, unit, or condition |
| Complete flows | Happy-path-only requirements, undefined errors and corner cases |
| Scope honesty | Feature lists with no stated exclusions |
| Dependencies & assumptions | Missing DCA sections, ownerless dependencies, beliefs stated as facts |
| Validation evidence | "Validated with users" claims with no method or linked findings |
| Audience fit | One-team tunnel vision, wrong format for the reader |
How the skill works
1
Identify the document
The skill confirms whether it audits a PRD, functional spec, or technical spec, and its stage.
2
Walk the checklist
It tests every requirement for testability and checks for missing sections, flows, and exclusions.
3
Deliver ranked rewrites
It rates findings by build damage and ends with the top three fixes, each with a rewrite.
Try it with
Related skills
1.0.0
MIT
Product Management
Audits
