The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests potentially untrusted evidence and logs while maintaining the capability to write reports. However, the design includes robust instructions that mitigate these risks.
Ingestion points: The skill uses Read, Glob, and Grep to analyze clinical recording artifacts and implementation evidence as described in the Instructions section of SKILL.md.
Boundary markers: The skill includes a dedicated Error Handling table that mandates an immediate stop to data collection if patient content is detected and requires the use of bounded evidence.
Capability inventory: The skill is granted access to Read, Glob, Grep, WebFetch, Write, and Edit tools.
Sanitization: The instructions explicitly require the production of "redacted escalation receipts" and strictly forbid replaying encounters or opening clinical notes during the diagnostic process.

abridge-common-errors