claude-adversarial-reviewlisted

# Adversarial review via headless Claude (`claude -p`) Claude as the review leg of the cross-family workflow. Use it when the implementer was GPT, Gemini, or a named free-pool model — never to review Claude's own work (a second Claude context is a fresh look, not model diversity; that is the one pairing that forfeits the point). ## Before the first run of a session Read **[`references/claude-runtime.md`](references/claude-runtime.md)** (same directory). It holds the family mechanics shared with `claude-implement` — invocation shapes, the MCP-stack trim, the patience calibration (five minutes of silence is normal, not a hang), the headless-plan-mode trap, and the instruction-layer inheritance property. This file assumes them and covers only the review role. ## Establish an immutable review target **One frozen directory per reviewer, at the exact commit, that nothing else touches — no lead activity inside it.** Not "whenever possible": a reviewer reads the WORKING TREE, not your commit. Measured — a round ran mutation testing in the same worktree mid-review and the reviewer opened a CRITICAL on a mutated, non-compiling file it was never meant to see. Freeze it with the suite's tested helpers instead of hand-rolling the shell; every bug ever found in this step was in a hand-rolled copy. The suite-root resolver and both calls — `freeze-target.sh` to create it, `verify-target.sh` before AND after the run — are in [dev-lead Phase 2](../dev-lead/SKILL.md), and the reasoning is