The Agent Skills Directory

[COMMAND_EXECUTION]: The script collect-facts.py executes local audit commands, such as pytest and specific internal scripts (e.g., check-doc-links.py), using the subprocess.run method. These operations are restricted to internal repository paths and are required for the skill's auditing functionality.
[DYNAMIC_EXECUTION]: The skill implements security-conscious execution by redirecting Python bytecode to a randomized, private temporary directory outside the repository using the PYTHONPYCACHEPREFIX environment variable. This prevents the generation of artifacts within the audited repository and mitigates local cache poisoning risks.
[DATA_EXPOSURE]: The audit process reads structural metadata and documentation content from repository files, such as SKILL.md and contexts/AGENTS.md, to verify line limits and formatting. The logic is purely local; there is no evidence of network exfiltration or unauthorized access to sensitive credentials like SSH keys or environment secrets.

spec-evolution