The Agent Skills Directory

[PROMPT_INJECTION]: The analyzer detected strings typically associated with prompt injection (e.g., 'ignore previous instructions') within references/evals.yaml. These were confirmed to be part of evaluation test cases designed to verify the security and resilience of other skills, which is a benign and intended use of the framework.\n- [COMMAND_EXECUTION]: Multiple scripts, including run_skill_evals.py, skill_gate.py, and quick_validate.py, use os.execve to re-execute themselves with a specific Python interpreter if dependencies like PyYAML are missing. The execution path is restricted to a specific virtual environment directory in the user's home folder. Other command executions use subprocess.run with argument lists to invoke linters and validators, which is standard behavior for an automated build tool.\n- [DYNAMIC_EXECUTION]: The use of os.execve for interpreter re-execution is localized and used for environment setup. No arbitrary or untrusted code execution was observed. The skill's primary workflow emphasizes 'fail-fast' validation and human-in-the-loop review before any significant changes are applied.

skill-builder