The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core documentation generation logic.
Ingestion points: The scripts/init_docs.py script reads and extracts data from project-level files including package.json, pyproject.toml, Cargo.toml, go.mod, and README.md (specifically via detect_project_type and extract_readme_summary).
Boundary markers: The generated AGENTS.md file, which is intended to serve as a primary instruction source for AI agents, incorporates this extracted content into markdown sections but fails to use specific security delimiters or 'ignore instructions' warnings around this untrusted content.
Capability inventory: The skill possesses the capability to modify project documentation (CLAUDE.md, AGENTS.md) and execute system commands (tree). Instructions promoted into AGENTS.md could potentially influence future agent actions in the project context.
Sanitization: There is no evidence of sanitization, validation, or escaping of the content extracted from external project files before it is written into the documentation.
[COMMAND_EXECUTION]: The skill executes the system tree command to generate directory structure visualizations.
Evidence: Found in scripts/init_docs.py and scripts/validate_docs.py using subprocess.run calls to the tree binary.
Context: The execution is performed safely using a list of arguments and the default shell=False, which prevents command injection from the path arguments. The execution is limited to a specific utility for documentation purposes.

doc-claude-md