The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents a multi-agent orchestration workflow where a coordinator agent dispatches task specifications to worker agents using the orca orchestration dispatch --inject command. This creates a surface for indirect prompt injection where workers ingest and act upon instructions provided by other agents.
Ingestion points: Worker agents receive task instructions and preambles via the terminal input managed by orca orchestration dispatch (SKILL.md).
Boundary markers: The documentation refers to "preambles" and "TASK blocks" to structure the data, but does not implement strict sanitization or isolation protocols.
Capability inventory: Agents can execute arbitrary commands, create new terminals, and modify files using the Orca CLI tools.
Sanitization: There are no specific instructions for sanitizing or validating external task specifications before injection.
[COMMAND_EXECUTION]: The skill relies on the orca CLI to perform actions such as terminal create --command and terminal send, which allow for the execution of arbitrary shell commands in managed agent sessions.

orchestration