The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process Python source code provided by users, which represents an untrusted data source. This creates an attack surface where instructions hidden in code comments or docstrings could attempt to manipulate the agent during analysis or rewriting.
Ingestion points: User-provided Python files or code snippets targeted for auditing or comment improvement (SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or "ignore" instructions to separate the target code from the agent's operational logic.
Capability inventory: The skill possesses the capability to rewrite code prose, docstrings, and comments based on its analysis of the input data.
Sanitization: There is no evidence of sanitization or filtering applied to the input source code to detect or neutralize embedded natural language instructions.
[EXTERNAL_DOWNLOADS]: The documentation references several external Python utilities for linting and documentation coverage.
Evidence: References to ruff, pydocstyle, and interrogate are found in references/docstring-styles.md.
Context: These are well-known, established tools within the Python community used for standard development workflows.
[COMMAND_EXECUTION]: The reference documentation includes examples of shell commands for installing and running development tools.
Evidence: Command examples such as pip install pydocstyle, ruff ., and interrogate -v src/ are provided in references/docstring-styles.md.
Context: These commands are provided for educational purposes within the documentation and are not executed automatically by the skill's operational logic.

python-comments