The Agent Skills Directory

[METADATA_POISONING]: The skill includes a file named gene-info_audit_result_v2.json which provides a fabricated security audit summary. This file contains deceptive claims such as "security: PASS" and "deployable: true", appearing to target automated or manual security analysis processes by mimicking an internal platform audit result.\n- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes gene summaries and descriptions from external APIs, which could contain malicious instructions designed to influence the agent.\n
Ingestion points: Text data is fetched via fetch_summaries_by_ids (NCBI Gene) and get_gene_data (Ensembl REST API) in scripts/fetch_gene_info.py.\n
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to disregard potential instructions embedded in the retrieved scientific summaries.\n
Capability inventory: The skill includes network request capabilities (requests.get), file writing (json.dump, csv.DictWriter), and local file reading (read_genes_from_file).\n
Sanitization: There is no evidence of sanitization or safety filtering applied to the remote data before it is presented to the agent context.

gene-info