The Agent Skills Directory

[SAFE]: The skill operates as a deterministic 'Stop hook' for the Claude Code agent environment. The Python script (quality-gate.py) uses shutil.disk_usage and os.path.getmtime to verify environment health and file updates.
[COMMAND_EXECUTION]: The script uses re.findall on the session transcript to count occurrences of Edit or Write tool calls to determine task complexity. This is an internal heuristic and does not execute shell commands or user-supplied code.
[DATA_EXPOSURE]: The skill accesses project-specific metadata files (e.g., growth-log, decisions/log.md) within the ~/.claude/projects/ directory to check for recent updates. Access is restricted to the current project's context, following privacy boundaries established by the agent platform.
[DYNAMIC_CONTEXT_INJECTION]: While the skill instructions demonstrate how to configure a shell command hook in settings.json, it does not utilize the !command syntax within the SKILL.md file itself, and the referenced script is intended for local execution by the agent host.

delivery-gate