adversarial-review

Performs independent red-team verification of spec-driven changes before archiving.

Updated Nov 1, 2024
One-click install
npx skills add https://github.com/mlorentedev/dotfiles --skill adversarial-review-mlorentedev
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: adversarial-review
Source: https://github.com/mlorentedev/dotfiles/tree/main/harness/skills/adversarial-review
Command: npx skills add https://github.com/mlorentedev/dotfiles --skill adversarial-review-mlorentedev

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Spec-driven changes often get archived based on the implementer's own verification, which misses gaps, incorrect assumptions, and negative-path failures. This Skill provides an independent adversarial review pass that tries to refute acceptance criteria before a spec is archived, producing a persisted verdict that the archive gate requires. ## Core Features & Use Cases - Adversarial Verification: Reads the spec triad (proposal, tasks, verification) plus the PR diff and actively tries to break each acceptance criterion, hunting race conditions, authz gaps, and spec-vs-code mismatches. - Structured Findings: Classifies every finding by severity (Blocker/Major/Minor) and reality (REAL/THEORETICAL/SPECULATIVE), requires a named covering test or marks it UNTESTED, and grades six dimensions on an A-D rubric. - Gate-Integrated Verdict: Writes specs/<feature-id>/review.md with machine-readable frontmatter (verdict, reviewed_sha, reviewer) that dotf spec archive validates before allowing archival. - Use Case: After implementing feature AI-001 and before running /spec archive, launch dotf spec review AI-001 so an independent reviewer session red-teams the change and issues a PASS, PASS-WITH-GAPS, or FAIL verdict. ## Quick Start Ask the AI to run an adversarial review on feature AI-001 before archiving its spec.

Frequently Asked Questions about adversarial-review

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

FAQPage Schema
How do I run an adversarial review on a spec-driven change?

Run `dotf spec review <feature-id>` in a repo with a reviewer pool, or invoke `/adversarial-review <feature-id>` in a separate session from the implementer. The review reads the spec triad and diff, then writes its verdict to `specs/<feature-id>/review.md`.

What is the difference between PASS, PASS-WITH-GAPS, and FAIL verdicts?

PASS means no blockers or majors and rubric grades of B or above. PASS-WITH-GAPS applies when only minors exist, open majors are theoretical or speculative, or the rubric has a C. FAIL results from any blocker, any REAL major, or any rubric grade of D.

Can the same agent that implemented a change review it?

No. The skill explicitly forbids single-agent self-review because independence is the source of its value. The review must run in a different session or agent from the implementer, ideally drawn from the repo's declared reviewer pool.

When should I not use adversarial review?

Skip it during implementation, for trivial changes that bypassed spec-driven development per Skip rules, or when a fresh passing review already exists. A declared `review: waived` with a reason in proposal.md is the auditable alternative to skipping silently.

Why does the archive command refuse without a review.md file?

Since CLI-034, `dotf spec archive` requires a fresh, passing review.md as a pre-flight check. It validates the frontmatter verdict, the reviewed_sha against contract file changes, and the reviewer identity against the pool, so a chat-only verdict does not satisfy the gate.