The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in autogen_runner.py to call the openclaw command-line tool. This is the primary method for invoking agents and routing LLM calls. The implementation uses a list of arguments rather than a raw shell string, which is a standard security practice to prevent command injection.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface. The build_context function in autogen_runner.py aggregates user-provided tasks and responses from multiple agents into a single prompt for the next turn. While this is necessary for a debate workflow, it allows one agent's response to potentially influence or override the instructions of another if not carefully monitored.

autogen-collab