← ClaudeAtlas

adversarial-review-lenseslisted

Use when a reviewer agent must produce an adversarial verdict via 4 independent lenses and weighted synthesis (Genesis A7 pattern)
SebastienDegodez/skraft-plugin · ★ 4 · Code & Development · score 60
Install: claude install-skill SebastienDegodez/skraft-plugin
# Adversarial Review Lenses Reviewer agents use this procedure to produce an independent, defensible verdict on a phase artifact. Pattern derived from Genesis Step 7 (Adversarial Review). ## When to use Every SKRAFT phase reviewer (`backlog-discoverer-reviewer`, `backlog-planner-reviewer`, `solution-architect-reviewer`, `acceptance-designer-reviewer`, `software-engineer-reviewer`) invokes this skill once per review pass, after reading the upstream phase artifact(s) and the relevant `*-review-criteria` skill. The number of lenses actually executed is governed by the repo-wide depth tier (held in `skraft-config.json`; read with `config.mjs get --key depthTier`, or taken from the orchestrator's dispatch payload): | Depth tier | Lenses required | |---|---| | `basic` | 1 (Completeness only) | | `standard` | 2 (Completeness + Business Fit) | | `comprehensive` (default) | 4 (all lenses) | | `custom` | as configured in `userPreferences.customDepth.reviewerLenses`, minimum 1 | ## The four lenses Each lens is executed **in isolation**. The reviewer must not let observations from one lens influence another. Run them in this order, recording each lens's findings independently before computing the synthesis. ### Lens 1 — Completeness Does the artifact cover every input that fed into the phase? - Are all referenced issues, requirements, or upstream artifacts addressed? - Are there obvious gaps (missing scenarios, missing components, missing acceptance criteria)? - Are the mandato