The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run across several files (create_agent.py, update_agent.py, run_all_tests.py, and test_integration.py) to execute local utility scripts.
The implementation uses list-based arguments and specifies sys.executable, which is a secure way to invoke sub-processes without invoking a shell, thereby preventing traditional command injection vulnerabilities.
The usage is confined to executing the skill's own scripts for validation and testing purposes.
[PROMPT_INJECTION]: The update_agent.py and create_agent.py scripts facilitate an indirect prompt injection surface by allowing users to input changelog descriptions that are then written into the 'Version History' section of agent definition files.
Ingestion points: The --changelog command-line argument and interactive text prompts for changelog entries.
Boundary markers: The input is formatted into Markdown bullet points, but there are no explicit delimiters or instructions to the LLM to ignore potentially malicious instructions embedded in the history section.
Capability inventory: The agents managed by this skill are typically granted significant capabilities, including the Read, Write, Edit, and Bash tools.
Sanitization: The scripts perform no sanitization of the input text, meaning malicious instructions could be persisted into the agent's long-term configuration.

agent-creation-update