The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages the automated installation of necessary third-party libraries during the bootstrap phase.\n
Evidence: _bootstrap/install.py utilizes pip install to download and install tree-sitter and its specific language grammars (tree-sitter-typescript, tree-sitter-javascript) from the Python Package Index (PyPI).\n- [COMMAND_EXECUTION]: The skill utilizes several system-level execution patterns for configuration management and code analysis.\n
Evidence: bootstrap.py employs os.execv for process replacement, ensuring the skill's setup runs under a compatible Python version.\n
Evidence: Multiple components, including _bootstrap/readiness.py, hooks/session-health-check.py, and hooks/_comment_judge.py, use subprocess.run to call git, python, pip, and the claude CLI.\n
Evidence: _bootstrap/permissions.py programmatically adds Bash allow-rules to the agent's settings.local.json to pre-approve the execution of its own linter and installer scripts, avoiding manual user prompts for these specific commands.\n- [DATA_EXFILTRATION]: The skill facilitates the evaluation of local workspace content by an external model service.\n
Evidence: hooks/judge-comments.py extracts source code comments from files modified in a session and transmits them to an Anthropic model via the claude CLI to judge their adherence to prose standards.\n- [SAFE]: All identified activities are consistent with the skill's primary and documented purpose as a coding standards enforcement tool.\n
Evidence: External downloads and tool references target established and trusted organizations and registries.\n
Evidence: The separate model calls used for comment evaluation are strictly sandboxed with tools disabled and a single-turn limit.

coding-standards