nda-triage-anthropic
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data in the form of NDA document text.
- Ingestion points: The agent reads legal text provided by the user in SKILL.md.
- Boundary markers: There are no explicit instructions or delimiters defined to separate the data from the agent's operating instructions.
- Capability inventory: The skill lacks any tool definitions, scripts, file-system access, or network permissions.
- Sanitization: No sanitization or filtering of the input text is performed.
- Given the absence of executable capabilities, the potential impact of an injection is restricted to informational deception rather than technical exploitation.
- [NO_CODE]: The skill contains no executable scripts, binaries, or tool configurations, adhering to a minimal attack surface and avoiding risks associated with remote code execution or persistence.
Audit Metadata