brida-reflex-meeting-utterance-triage
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions mandate checking live vendor resources (/reflex/llms.txt, /reflex/agent.md, and /reflex/agent.json) to determine current API and SDK availability. These resources are associated with the vendor, brida-ai.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external meeting utterances, creating a surface for potential indirect prompt injection. Mitigations include: 1. Ingestion points: Data is ingested via the utterance field in the custom-reflex.json contract. 2. Boundary markers: The agent is restricted by a declarative_policy that limits outputs to a predefined set of semantic branches. 3. Capability inventory: Capabilities are limited to semantic classification; the skill explicitly prohibits performing protected actions based on results. 4. Sanitization: The skill requires normalization of the state and instructs the agent to judge only the supplied utterance without inventing context.
Audit Metadata