The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured output from the external BioMCP CLI tool, creating an attack surface for indirect prompt injection if the source data contains malicious instructions designed to influence agent behavior.
Ingestion points: toolkit/scripts/reconcile.py reads and parses data from captures.json (BioMCP output).
Boundary markers: toolkit/README.md explicitly instructs the agent to preserve but ignore cached enrichment or provider instructions for the purpose of command choice.
Capability inventory: toolkit/scripts/reconcile.py performs file system reads and writes limited to a newly created output directory.
Sanitization: The script implements strict JSON schema validation, type enforcement (e.g., isinstance checks), and robust path traversal protection in the safe_path and capture_path functions.
[EXTERNAL_DOWNLOADS]: The documentation references and facilitates the installation of the biomcp-cli Python package (version 0.9.3) from PyPI and references the GenomOncology GitHub repository.

omh-biomedical-evidence