The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run within scripts/_skill_cli.py to execute project-related commands such as npm, pip, and poetry. The implementation uses shell=False and resolves executables via shutil.which, which is a security best practice to prevent command injection.
[EXTERNAL_DOWNLOADS]: The ensure_npm and ensure_python scripts facilitate the installation of dependencies from standard registries (NPM and PyPI). These are typical developer workflows and target official package managers.
[INDIRECT_PROMPT_INJECTION]: The skill's auditing and stack detection logic (audit_scaffold.py, _project_detect.py) reads metadata from project files such as package.json, pyproject.toml, and PRODUCT.md. While these files constitute an external data source that could theoretically contain instructions, the skill uses them strictly for pattern matching and JSON parsing to determine project configuration, presenting a minimal attack surface.
Ingestion points: Reads package.json, pyproject.toml, PRODUCT.md, and .heyeddi/stack.json in _project_detect.py and audit_scaffold.py.
Boundary markers: None explicitly defined for these files, as they are standard project manifests.
Capability inventory: The skill can write files to the project directory, install dependencies via npm/pip, and run tests via vitest/pytest.
Sanitization: Uses standard json.loads for parsing metadata and basic string matching for stack detection.

project-engineering