adversarial-reviewing

Runs multi-agent adversarial code review via a deterministic FSM orchestrator. Dispatches specialist sub-agents through self-refinement, challenge, red team audit, and resolution phases. Use when the user asks for adversarial review, security review, code review, or multi-agent review.

Adversarial Review

This skill runs a multi-agent security review via a Python orchestrator.

Quick start

Run exactly this command:

cd ${CLAUDE_SKILL_DIR} && python3 -m scripts.orchestrator run-all $ARGUMENTS

This runs the full review pipeline (init, confirm, self-refinement, challenge, red-team audit, report). No further action needed.

The command will take 30-90 minutes depending on repo size. Wait for it to complete. Use timeout: 600000 on the Bash tool call.

Output

When complete, the command prints JSON with {"status": "done"}. Review artifacts are in ${CLAUDE_SKILL_DIR}/artifacts/.

Legacy workflow

If run-all is not available, fall back to the step-by-step protocol:

cd ${CLAUDE_SKILL_DIR} && python3 -m scripts.orchestrator init $ARGUMENTS

Then follow the dispatch.json loop (confirm, dispatch agents, next) until done=true.

Error handling

If the command fails, report the error and exit. Do not retry. Do not re-initialize. Do not improvise recovery.