The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and other shell commands (e.g., git, grep, mkdir) to gather project metadata and execute the internal metrics script. This is restricted to read-only operations on the local file system and repository history.
[DYNAMIC_EXECUTION]: The skill extracts a Python script (graph_metrics.py) from its own markdown documentation and executes it using the python3 interpreter. This script is restricted to the standard library and includes a self-integrity check (SELF_DIGEST) to prevent unauthorized modifications to the analysis logic.
[SAFE_PRACTICE]: The skill explicitly enforces a 'Read-Only Discipline', stating it must not write to, edit, or create files inside the audited repository, and limits output to a dedicated reports/ directory.
[SAFE_PRACTICE]: Sensitive configuration files like .env and CODEOWNERS are read to understand the project structure and ownership without evidence of exfiltration to external domains.

architecture-audit