OpenAI's Codex Security CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. npm: https://www.npmjs.com/package/@openai/codex-security
Scan Runtime
Use the current tool surface for Standard and diff scan delegation. If delegation tools are deferred and tool_search is available, discover them once before deciding that workers are unavailable. Use only successful spawn calls and their returned worker IDs as evidence that workers started. Configured capacity is a maximum, not a required number of workers.
When delegation is unavailable or a worker cannot start, continue the assigned review in the parent. Account for every authorized file and security question, and report actual coverage gaps. Do not inspect or edit persistent Codex configuration to enable optional workers or goal tools. Apply relevant SECURITY.md guidance and use available goal tools when the task authorizes them; their absence does not block source review.
For a host-backed diff scan, use its authoritative scan context and advance to threat_model before starting architecture mapping. The host still validates ownership, target, scope, authentication, and sandbox permissions. Deep Scan's coordinator owns its worker setup and progress; it does not depend on this thread's delegation tools.
A scan from an older workbench may have no recorded owner. The workbench can still display it. A conversation can continue it once the host attaches its thread.
Daybreak Access Advisory
For Standard scans, follow the launch instructions for the optional account check. For a diff scan, read access-advisory.md before the first substantive scan action. It does not gate the scan or change runtime setup.