embodiment-description
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text files to generate patent descriptions while operating in an environment with high-privilege execution capabilities.
- Ingestion Points: Reads
patent/INVENTION_DISCLOSURE.mdandpatent/CLAIMS.mdfiles as part of the analysis workflow. - Boundary Markers: Absent. The instruction set does not supply clear structural delimiters or negative constraints to prevent processing embedded instructions within the source text files.
- Capability Inventory: Frontmatter allows the use of
Bash(*),Read,Write,Edit,Grep, andGlobtools. - Sanitization: Absent. No explicit validation or filtering mechanisms are declared for the handled markdown documents.
Audit Metadata