The Agent Skills Directory

[SAFE]: The skill manages state using a dedicated project directory (opc-doc/), which prevents unauthorized access to sensitive system files.
[SAFE]: Explicit 'Human-in-the-loop' protocols require user confirmation before the agent writes formal conclusions or transitions between workflow stages.
[INDIRECT_PROMPT_INJECTION]: The skill reads from local state files to restore session context, which constitutes an injection surface if the environment is shared or untrusted. 1. Ingestion points: Files in opc-doc/state/ and opc-doc/outputs/. 2. Boundary markers: Not explicitly defined in the file-reading logic. 3. Capability inventory: File system read/write via platform tools and the ability to coordinate transitions between sub-skills. 4. Sanitization: No validation or sanitization is applied to the content of restored state files.

opc-orchestrator