The Agent Skills Directory

[SAFE]: The skill implements a legitimate scientific workflow for drug discovery. All external data requests are directed to official, well-known bioinformatics API endpoints (ebi.ac.uk, ncbi.nlm.nih.gov, opentargets.org, proteinatlas.org, and uniprot.org).
[COMMAND_EXECUTION]: The scripts/orchestrate.py file uses subprocess.run to execute internal fetcher scripts in parallel. This is a standard orchestration pattern and is implemented safely using command lists rather than shell strings, which prevents shell injection attacks.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided gene lists (CSV or text) to generate a report that is later interpreted by an AI agent. While this constitutes an attack surface for untrusted data, the risk is negligible as the input is strictly treated as query parameters for scientific databases and the final rationales are grounded in the retrieved structured data.
Ingestion points: orchestrate.py reads gene symbols from the --input file; aggregate.py reads expression data from the --input-csv file.
Boundary markers: Absent for the input data processing, but the data is constrained by API validation.
Capability inventory: Subprocess execution (orchestrate.py), network API access (fetch_*.py), and file system writes (aggregate.py).
Sanitization: Gene symbols are validated against scientific databases; the report template encourages factual grounding and discourages hallucination.

target-prioritization