The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several scripts that interact with the system via shell commands or subprocesses. Specifically, scripts/make_contact_sheet.py uses subprocess.run to invoke the pdftoppm utility. This implementation uses a list of arguments rather than a raw shell string, which significantly reduces the risk of command injection.
[DYNAMIC_EXECUTION]: The scripts/pdf_qa.sh script executes a small Python snippet using python3 -c to perform CJK character detection. This execution is limited to a hardcoded string and is used for metadata validation purposes rather than processing external code.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection as it processes external files and data blocks (PDF content and YAML logs).
Ingestion points: pdftotext in pdf_qa.sh reads PDF body text, and yaml.safe_load in lint_run.py parses structured blocks.
Boundary markers: The workflow relies on strict YAML schemas and documented handoff formats to maintain context boundaries.
Capability inventory: The skill can execute local binary tools (poppler-utils) and filesystem writes within a specific directory structure.
Sanitization: scripts/lint_run.py implements a validation function _path_in_work that normalizes file paths and rejects any absolute paths or '..' parent directory escapes, ensuring file operations are sandboxed within the work/ directory.

mathodology-award-gates