faffter-noon-spec-review

Reviews software specifications through four lenses and emits a structured verdict contract.

3|Updated Apr 16, 2026
One-click install
npx skills add https://github.com/shftwst/faff --skill faffter-noon-spec-review-shftwst
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: faffter-noon-spec-review
Source: https://github.com/shftwst/faff/tree/main/plugin/skills/faffter-noon-spec-review
Command: npx skills add https://github.com/shftwst/faff --skill faffter-noon-spec-review-shftwst

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Specs often ship with unsound designs, security gaps, wrong-sized scope, or unverifiable acceptance criteria that only surface after code is written. This Skill challenges the approach while the work is still just a spec, producing a single founded verdict that routes the work forward, back for revision, or to a human. ## Core Features & Use Cases - Four-lens single-pass review: Walks architectural, infosec, methodology, and QA lenses as one checklist over a freshly-rated spec, emitting at most one objection per lens with claim, evidence, and predicted consequence. - Deterministic verdict roll-up: Maps objection severities to approve, revise, reject-approach, or needs-human verdicts, and emits a faff-contract:spec-review-verdict JSON block validated by the faff contract CLI. - Level-aware depth: Ships the L1–L3 single-pass form while leaving the slot open for an L4 adversarial per-lens occupant; lens selection by change-surface acts as a cost gate. - Use Case: After faff-prep produces and confidence-rates a spec, this reviewer runs once before promote-to-Todo, flags a blocking architectural objection with a spec anchor, and routes the spec back for re-specification instead of letting a flawed design reach implementation. ## Quick Start Ask the agent to review the freshly-rated spec through the spec_review slot and emit a spec-review-verdict contract block before promoting the work item to Todo.

Frequently Asked Questions about faffter-noon-spec-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a software spec before implementation with an AI agent?▼

Run the spec_review slot after the spec is confidence-rated and before promoting to Todo. The reviewer walks four lenses as a single checklist pass, gathers objections with claim, evidence, and predicted consequence, and emits one verdict contract block.

What review lenses does the spec reviewer apply?▼

It applies architectural, infosec, methodology, and QA lenses. Architectural checks design soundness and ADR consistency, infosec checks threat surface, methodology consumes the attached critique without recomputing scope, and QA checks verifiability of done criteria.

How are spec review verdicts decided from objections?▼

A deterministic roll-up maps severities to verdicts: no objections means approve, an architectural or methodology blocker means reject-approach, an infosec blocker means needs-human, and a QA blocker or any major/minor findings mean revise.

Can the spec reviewer run with fewer than four lenses?▼

Yes, the consumer's change-surface lens-selection step passes the lens-set to fire, and only low-risk lenses are skippable. Infosec and QA always fire at L1–L3, and with no lens-set passed all four fire by default.

What happens when no methodology critique is attached to the spec?▼

The methodology lens degrades to no signal available, emits no methodology objection, and notes the gap. It never falls back to recomputing value, scope, or risk on its own.

When should I use the adversarial spec reviewer instead of the single-pass one?▼

Use the adversarial per-lens refuter occupant at L4, where independent refuters attack the spec from each angle gated on majority and severity. The single-pass reviewer covers L1–L3 where one checklist pass is the cost-stable depth.