patent-claims-analyzer
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied patent claims as input for analysis. There are no explicit boundary markers or sanitization steps mentioned for this untrusted data. If a malicious input contains instructions disguised as patent text, the agent might inadvertently execute them.
- Ingestion points: SKILL.md (via the claims_text variable or file input).
- Boundary markers: Absent in the provided instructions.
- Capability inventory: Tool access includes Bash, Read, and Write operations.
- Sanitization: No evidence of input validation or escaping for the processed text.
- [COMMAND_EXECUTION]: The skill instructions involve executing Python code through the Bash tool to import and run a local claims_analyzer.py module. While the module is local, this capability requires the agent to interact with the shell environment.
Audit Metadata