The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/_skill_cli.py defines a run_command function that uses subprocess.run. The implementation follows best practices for safety, such as setting shell=False and resolving the executable path via shutil.which to prevent command injection. Furthermore, this utility is not actively invoked by the validation script.
[INDIRECT_PROMPT_INJECTION]: The skill includes a validation tool (scripts/validate_composable.py) that reads and analyzes code from the user's workspace. While this is an ingestion point for untrusted data, the script uses non-executable regular expressions for pattern matching and does not interpolate the content into agent prompts or execute it, significantly minimizing risk.
Ingestion points: scripts/validate_composable.py reads file content using target.read_text().
Boundary markers: The script does not use specific boundary markers as it processes the file content via static regex analysis.
Capability inventory: The script is limited to reading files and emitting JSON results to stdout; it lacks network access or file-writing capabilities.
Sanitization: The script performs passive regex matching and does not interpret input as instructions.

composable-patterns