The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md explicitly direct the agent to 'write and run Python code via Bash' for statistical analysis and data processing. This is a standard functional requirement for data-intensive agent tasks and does not involve malicious intent.
[EXTERNAL_DOWNLOADS]: The skill fetches data from numerous well-known and trusted scientific services including ClinicalTrials.gov, OpenTargets, CIViC, ChEMBL, PubMed, and FDA databases. These sources are reputable and required for the primary function of clinical trial matching.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, third-party content such as clinical trial descriptions and eligibility criteria.
Ingestion points: Data is ingested via various tool calls to ClinicalTrials.gov, PubMed, and OpenTargets as described in TRIAL_SEARCH_PATTERNS.md.
Boundary markers: The instructions do not define specific boundary markers for external text processing, although the final report follows a structured REPORT_TEMPLATE.md.
Capability inventory: The skill has access to bash command execution, python scripting, and local file writing for report generation as specified in SKILL.md and MATCHING_ALGORITHMS.md.
Sanitization: There is no evidence of specific input sanitization for external data before it is parsed by the agent.

tooluniverse-clinical-trial-matching