The Agent Skills Directory

[SAFE]: The skill is a legitimate research tool authored by Zaoqu-Liu, designed for the computational validation of drug target hypotheses. It employs a structured methodology with clear phases, identity resolution, and quantitative scoring metrics.
[EXTERNAL_DOWNLOADS]: The skill retrieves biological and medical data from a extensive list of well-known, public, and trusted scientific resources, including OpenTargets, ChEMBL, UniProt, Ensembl, gnomAD, the FDA, and PubMed. These downloads are integral to the skill's primary function and do not involve sensitive user data or executable code from untrusted sources.
[COMMAND_EXECUTION]: The workflow uses a library of specialized tools (tu.tools) to query APIs and process structured scientific data. All command examples are restricted to biological research tasks and use verified parameters to interact with legitimate data providers.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests and processes untrusted text from external sources such as PubMed abstracts, EuropePMC articles, and OpenAlex metadata.
Ingestion points: External data enters the agent context through tools like PubMed_search_articles, EuropePMC_search_articles, and OpenAlex_search_works as seen in SKILL.md and REFERENCE.md.
Boundary markers: The instructions lack explicit delimiters or warnings to treat external content as untrusted data, which could allow instructions embedded in literature to influence the agent's behavior.
Capability inventory: The agent has access to over 60 tools, file system writes (for report generation), and complex reasoning capabilities.
Sanitization: There is no mention of text sanitization or filtering to mitigate the risk of instructions hidden within retrieved scientific literature.

tooluniverse-drug-target-validation