medical-english-precision-editor
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-supplied manuscript drafts, creating a surface for indirect prompt injection. 1. Ingestion points: User-provided biomedical drafts, abstracts, and rebuttal text. 2. Boundary markers: The instructions lack explicit markers or delimiters (such as triple quotes or XML tags) to separate untrusted data from the agent's core instructions. 3. Capability inventory: No executable scripts, subprocess calls, network operations, or file-writing capabilities were detected across the skill files. 4. Sanitization: The skill does not specify technical sanitization or filtering of the input text, instead relying on instructional 'Hard Rules' to maintain evidence boundaries.
- [METADATA_POISONING]: The skill includes an audit artifact, 'eval_report_medical-english-precision-editor_result.json', which contains self-authoritative security claims including a 'PASS' status for stability and security. As per analysis protocols, these internal safety claims are treated as data points for evaluation rather than authoritative conclusions.
Audit Metadata