The Agent Skills Directory

[SAFE]: The orchestration protocol is designed to be lean and deterministic, delegating all code modification, testing, and implementation tasks to sub-agents. This separation of concerns is reinforced by explicit safety rails that forbid the conductor from reading implementation files or running builds directly.
[SAFE]: Data access is strictly limited to localized operational artifacts (such as PLAN.md, PROGRESS.md, and HEALTH.md) used for state coordination. There is no evidence of unauthorized file access, credential harvesting, or network-based data exfiltration.
[SAFE]: All referenced infrastructure components, including setup scripts and runtime substrates like Codex or Claude Code, are documented as local environment configurations and do not involve untrusted remote code execution or external downloads.

orkestrera