A comment isn't a fix
Peer review usually catches that a requirement is weak. It rarely produces a rewrite on the spot, so the fix waits for another cycle, or doesn't happen. Scoring every requirement and pairing each weak one with a specific rewrite collapses that review-and-rework loop into one pass.
What it does
- Checks five things per requirement. Ambiguity, missing actor, missing exception behaviour, compound "and/or" statements, and whether it's actually measurable.
- Scores 0–100, justified. Never a perfect score without a reason attached.
- Rewrites only what needs it. Under 70 gets a targeted rewrite; anything passing is left alone.
- Names the specific problem. Not "needs work" — the actual fault, so the rewrite is traceable to the fault it fixes.
- Summarised as one table. Requirement ID, score, problems, rewrite where applicable — easy to drop into a review pack.
Example
A constructed example, not real client requirements.
BR-04: The system should be fast and handle errors appropriately when users submit forms.
BR-04 | Score: 28/100 Problems: "fast" is not measurable; "appropriately" is not testable; compound requirement bundles performance and error handling. Rewrite (split into two): BR-04a: The system shall return form submission results within 2 seconds under normal load. BR-04b: The system shall display a field-specific error message when a required form field is left blank.
How it works with Claude
A one-off 10-minute install into your Claude account.
Paste in your requirements list.
Get scores, named problems and rewrites for the weak ones.
FAQs
What makes a requirement bad quality?
Ambiguity, a missing actor, missing exception behaviour, compound and/or statements that are really two requirements, and language that can't be objectively tested as met or not met.
Does it rewrite every requirement?
No, only the ones that score under a quality threshold. Requirements that already pass are left as written, since rewriting something that already works just introduces review overhead for no benefit.