The Agent Skills Directory

[COMMAND_EXECUTION]: Executes standard development tools such as git and gh (GitHub CLI) to retrieve status and history for context recovery. Analysis confirms these are read-only operations for gathering project metadata.
[COMMAND_EXECUTION]: Invokes a bundled Python script (scripts/passport.py) for state validation. The script is verified to use secure data handling practices, including yaml.safe_load() in its YAML parser.
[PROMPT_INJECTION]: The skill manages an attack surface for indirect prompt injection by ingesting external data such as Git logs, GitHub CI status, and chat transcripts. This risk is effectively mitigated by a mandatory evidence chain: (1) Ingestion points: Context is gathered from Git history and CI logs; (2) Boundary markers: Structured .light/passport.yaml files and handoff cards delineate state data; (3) Capability inventory: Limited to filesystem writes in the .light/ directory and CLI execution for status checks; (4) Sanitization: Mandatory schema validation is performed via the provided Python utility before state is updated.
[EXTERNAL_DOWNLOADS]: Communicates with GitHub services to fetch status updates via the gh tool. This interaction targets a well-known technology service and is considered a safe development practice.

light-orchestrator