The Agent Skills Directory

[COMMAND_EXECUTION]: The Python scripts detect_stack.py, scan_repo.py, and map_structure.py utilize subprocess.run to invoke git for retrieving repository file listings and diff information. These calls are limited to standard, non-destructive git queries used to inform the audit process.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes project source files to identify code smells and structural issues. While this exposes the agent to data from the local filesystem, the skill mitigates risk by utilizing deterministic scripts and providing structured feedback through generated Markdown and JSON documentation.
[DYNAMIC_CONTEXT_INJECTION]: The skill includes optional hooks for the Claude Code host in assets/hooks/claude-settings.json. These hooks execute shell commands at session start and after edits to automatically load project context from the .clean/ directory and perform architectural validation using the skill's local logic.
[SAFE]: No malicious behaviors such as obfuscation, credential harvesting, or remote code execution were detected. The skill adheres to its documented constraints by using the Python standard library exclusively and avoiding all network activity.

clean-code