Skip to content
Back to skills

Verdict Spec Review

ASecurity

Judge a specification, issue or PRD for testability before any code exists — inventory the requirements, find the untestable, contradictory, unbounded and silent ones with a verbatim quote each, and rewrite the core as Given/When/Then acceptance criteria. Use when a spec is about to be implemented.

  • 2 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 24, 2026
ai-agents

Works with

  • claude code

Security analysis

A100/100

Scanned October 4, 2026

npx -y skills add ArtJack/verdict --skill verdict-spec-review --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Verdict Spec Review?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Verdict Spec Review
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/artjack-verdict-spec-review/badge)](https://www.skillsdirectory.com/skills/artjack-verdict-spec-review)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: verdict-spec-review
user-invocable: false
description: Judge a specification, issue or PRD for testability before any code exists — inventory the requirements, find the untestable, contradictory, unbounded and silent ones with a verbatim quote each, and rewrite the core as Given/When/Then acceptance criteria. Use when a spec is about to be implemented.
---

# Verdict: spec review, before code

The cheapest defects to fix are the ones caught before code exists. No code is required and
none is judged.

**In Claude Code with the Verdict plugin, use `/verdict:spec <path>`.** This skill is the
same procedure for agents that cannot run it.

## Procedure

1. **Inventory the requirements.** Number every testable claim (R-1, R-2 …). A sentence
   that cannot fail a test is not a requirement — that is itself a finding.
2. **File findings**, each with a severity and a verbatim quote of the offending line:
   - **Untestable / unmeasurable** — "fast", "reliable", "user-friendly", "handles all edge
     cases" without a number or an observable behaviour.
   - **Contradictions** — requirements that cannot both hold; quote both lines.
   - **Undefined boundaries** — behaviour exactly at the limit, rounding direction,
     inclusive vs exclusive ranges, empty and zero cases, time zones, calendar vs business
     days. Name the boundary each ambiguity hides.
   - **Silent gaps** — failure paths, permission cases, partial or concurrent operations
     the spec never mentions.
   - **Conflicts with recorded history** — a changelog, ADR or earlier decision that
     contradicts the spec; cite it.
3. **Write acceptance criteria.** Rewrite the core requirements as Given/When/Then, precise
   enough that the criterion *is* the test — implementer-ready.
4. **Park the questions only a person can settle** (which reading of an ambiguity is the
   spec?) as `questions` in the judgment rather than deciding them yourself; the
   maintainer answers with `verdict-answer`, and the next run reads the answer.
5. **Verdict on the spec:** `pass` means implementable and testable as written; anything
   less says exactly what change would earn it.

Spec findings are real findings: file them through the harness like any other
(`verdict-facts` → `<qa-root>/findings/<ID>.json`, evidence citing `SPEC.md:line`,
`failure_classification: null` → `verdict-finalize`), so they age, resolve and regress as
the spec is revised.

Verdict reports and specifies; it never fixes. Route the fix to the implementer, then
verify it (`verdict-verify-fix`).

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.