Subagent Orchestration

Use this skill after engagement-scope has established the authorized target, scope evidence, impact tolerance, and deliverable. This skill plans and coordinates subagents; it does not replace the technical skill that owns the actual evidence, proof, patch, or report.

Codex only spawns subagents when the user explicitly asks for them. Treat that as an intentional cost and latency boundary. Do not use this skill to justify silent fan-out for ordinary depth, thoroughness, or codebase exploration.

Staged Agent Model

Use a staged security workflow: prepare the target, scan with specialized auditors, validate with separate debaters, deduplicate equivalent findings, and prove candidates with dynamic or domain-specific proof helpers. The lesson is not "spawn many agents." The lesson is role separation, independent evidence, and a pipeline that turns candidates into proofed findings instead of a pile of noisy claims.