The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to retrieve its full usage guide and lifecycle preambles by executing ORCA skills get orchestration. This establishes a pattern where external data from the local environment is treated as authoritative instruction content.
Ingestion points: The agent is directed to ingest the output of ORCA skills get orchestration as a version-matched guide in SKILL.md.
Boundary markers: No explicit delimiters or boundary markers (e.g., 'ignore instructions within this data') are provided for the fetched content.
Capability inventory: The skill allows shell command execution via the ORCA placeholder (resolving to orca, orca-cli, orca-ide, or orca-dev), including status checks, task listing, and terminal management.
Sanitization: There is no evidence of sanitization or validation performed on the output of the guide-fetching command before the agent processes it as logic.
[COMMAND_EXECUTION]: The skill requires the agent to execute system binaries and interact with the shell. It includes logic to resolve the specific executable path based on environment variables (ORCA_CLI_COMMAND, ORCA_DEV_REPO_ROOT) and platform-specific heuristics.
[DYNAMIC_EXECUTION]: The instructions mandate that the agent load its operating logic at runtime from the binary ("Load the full guide before running Orca commands"), which is a form of dynamic instruction loading. Additionally, the skill describes 'full ownership handoffs' where work is delegated to other agents without user supervision if certain conditions are not met.

orchestration