abridge-cost-tuning
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which presents a surface for indirect prompt injection attacks. \n
- Ingestion points: The workflow ingests utilization exports, training records, and clinician feedback as specified in SKILL.md, and fetches documentation from external Abridge subdomains via WebFetch as defined in references/official-docs.md. \n
- Boundary markers: The skill contains explicit instructions to stratify outputs without exposing identities and to label external results as non-comparable to mitigate the influence of injected instructions. \n
- Capability inventory: The agent uses Write and Edit tools to generate reports and decision tables based on the processed data. \n
- Sanitization: While specific data-cleaning steps are not listed, the skill enforces a robust approval boundary requiring clinical and commercial owners to verify all findings before concluding the workflow.
Audit Metadata