adversarial-reviewerlisted

# Adversarial Reviewer This skill performs an explicit, read-only adversarial review of active task work. It checks whether the work is ready for Finalization. ## Independence This review MUST be performed by an **independent agent with a fresh context** where the environment supports it — dispatched as a subagent, not run inline in the working conversation — so the work is judged by a reviewer that did not write it. If no subagent or fresh-context capability is available, the main agent must not pretend the review was independent and must not silently skip it. Notify the user that independent review is unavailable and perform an explicitly marked inline adversarial review as the fallback. Record the limitation in the review report. Independence also depends on what the dispatch prompt contains. The invoker's dispatch must carry pointers to the artifacts — the brief, the progress file, the task delta, relevant General/Situational — and the current delta anchor, and nothing else (defined in `../../../ai-context/methodology/adversarial-review.md`). If the dispatch prompt instead included the implementing agent's own summary, conclusions, or argument for why the work is correct, note that as a Methodology Feedback Observation: the review may have been biased by the author's framing rather than independent of it. A dispatch naming no current delta anchor — a direct invocation — is missing its boundary: compute the anchor yourself by the standard recipe (defined in `../../../