The Agent Skills Directory

[SAFE]: The skill implements strong defensive instructions, explicitly prohibiting the exposure of credentials, passwords, session tokens, or Protected Health Information (PHI). It mandates the use of synthetic data for troubleshooting and redaction of sensitive data in any generated reports.
[SAFE]: Tool usage is restricted to legitimate administrative tasks. The instructions limit WebFetch strictly to official first-party documentation and restrict file modifications (Write, Edit) to user-requested deliverables at approved destinations.
[SAFE]: The skill maintains clear clinical safety boundaries by requiring an independent clinical fallback for urgent care needs and ensuring that all triage activities remain informational rather than diagnostic.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it fetches content from external OpenEvidence documentation and processes user-provided symptom data. However, the risk is minimized by instructions to use only first-party documentation and explicit requirements for human clinical review and governance oversight.
Ingestion points: External OpenEvidence documentation fetched via WebFetch and user-supplied symptom strings defined in SKILL.md examples.
Boundary markers: The instructions explicitly demand minimal inputs, the separation of facts from assumptions, and the use of authorized, synthetic data.
Capability inventory: Uses Read, Glob, Grep, WebFetch, Write, and Edit tools.
Sanitization: Instructions mandate the redaction of patient and credential data before any output is generated.

openevidence-rate-limits