el-red-reviewlisted

# el-red-review — Claude Code fulfilment of the EL adversarial challenger EL's adversarial Red Reviewer composes two challengers: 1. **Rule-based challenger** — ALWAYS runs, deterministic. Sets each finding's `red_review.status` during the run so the pipeline never blocks. 2. **LLM challenger** — runs in-process when `ANTHROPIC_API_KEY` is set. When the key is absent but EL is inside a Claude Code session (or the operator set `EL_RED_REVIEW_DEFER=1`), EL writes a request file at `reports/_red_review_request.json` instead of skipping the LLM. **You are that LLM challenger, via this skill.** The challenger's job is to **falsify** findings, not validate them. A stricter LLM verdict can lift a finding from `passed` to `challenged`; it never fake-passes. The merge is severity-max (`challenged > unresolved > passed`). ## When to fire - The user explicitly invokes `/el-red-review` (with or without a case-id). - A new `_red_review_request.json` appears under any `/opt/EL/cases/<id>/reports/` (e.g. after an `el investigate` run inside Claude Code with no API key). If the user hasn't asked and there are no pending requests on disk, do nothing — this skill is a worker, not a watchdog. ## Procedure 1. **Discover pending requests**. ```bash find /opt/EL/cases -maxdepth 3 -name "_red_review_request.json" -type f ``` If the user passed a case-id, scope to that case. If nothing matches, say "no pending red-review requests" and stop. 2. **Read the