The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources including PubMed, EuropePMC, and DailyMed through tools like PubMed_search_articles and DailyMed_parse_clinical_pharmacology. Because these tools retrieve potentially untrusted text (such as article abstracts or clinical narratives) that is then processed by the agent without explicit boundary markers or sanitization guidelines in the skill instructions, it creates a minor surface for indirect prompt injection.
[SAFE]: The skill uses established scientific and regulatory databases (e.g., ChEMBL, KEGG, Reactome, STRING) which are well-known services in the bioinformatics domain. The use of these sources is appropriate for the skill's stated purpose of pharmacological research.
[SAFE]: No evidence of hardcoded credentials, obfuscated code, or unauthorized network exfiltration was found. All network operations are conducted through defined tools targeting reputable APIs.
[SAFE]: The skill does not attempt to gain elevated privileges, modify system configuration files, or establish persistence mechanisms.

tooluniverse-drug-mechanism-research