negotiation-prep
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as counterparty messages and emails, which could potentially contain hidden instructions for the AI.
- Ingestion points: The agent reads counterparty messages in ANALYZE mode and context from external files or logs in the Context Loading step in SKILL.md.
- Boundary markers: No specific delimiters or ignore instructions are defined for the agent to use when ingesting these external texts.
- Capability inventory: The skill primarily performs text analysis and scoring; no dangerous capabilities like network access or command execution were found across the provided scripts and instructions.
- Sanitization: The instructions do not explicitly require sanitization of the external data before analysis.
- [SAFE]: No other security concerns were identified during the analysis. The skill follows its stated purpose and does not exhibit patterns of credentials exposure, persistence, obfuscation, or malicious dynamic execution.
Audit Metadata