Five-way parallel code audit. User pastes a Python file / function /
diff → orchestrator fans out 5 specialist sub-agents in ONE message,
each reviewing a DIFFERENT axis (static analysis, library API
correctness, standards/spec compliance, reproducibility, numerical
correctness). Every finding MUST cite an exact URL + a quoted
passage from the official documentation. Aggregator agent merges,
dedupes, ranks by severity. Code returns reviewed 4-5 times before
the user touches it.
Trigger when user says: "audit this code", "review this diff",
"debug this", "check my function", "find bugs", "/audit-code",
"/code-audit-fanout", "fanout audit", or pastes a code block and
asks …
