The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/inbox-loop.py executes shell commands using the subprocess module. Specifically, it uses subprocess.Popen to call the Orca CLI. The executable path is determined by the ORCA_CLI_COMMAND environment variable, which allows the execution of arbitrary commands if the environment is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function as a coordinator requires it to ingest data from various untrusted or external sources, including worker messages, pull request descriptions, and external issue trackers.
Ingestion points: External messages are retrieved and stored by the scripts/inbox-loop.py script as described in references/inbox-and-waiting.md. PR metadata and descriptions are processed during the gate acceptance phase in references/gate-acceptance.md.
Boundary markers: The skill uses structured templates for some communications, but it does not consistently implement strict delimiters or instructions to ignore embedded commands when processing content from workers or trackers.
Capability inventory: The skill possesses significant capabilities, including local command execution via the Python script, file system modifications through git merge operations, and network operations via git push/pull.
Sanitization: The provided files do not show explicit sanitization or escaping of content received from external workers before it is processed by the coordinator agent.

orchestra-auto-zh