brida-reflex-research-event-relevance
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external 'events' to determine their relevance to specific research questions. This creates a surface where untrusted data could attempt to influence agent behavior through embedded instructions.\n
- Ingestion points: The agent is instructed to 'locate the event/state that feeds this decision' (SKILL.md) and process state definitions in references/custom-reflex.json.\n
- Boundary markers: The skill establishes an 'Authority boundary' in SKILL.md and references/playbook.md, explicitly stating the reflex is recommendation-only and does not grant permission for protected actions.\n
- Capability inventory: The skill utilizes MCP, SDK, and REST integration surfaces for execution (SKILL.md).\n
- Sanitization: Instructions require the agent to 'Normalize only the bounded state required by the bundled contract' and 'Accept only declared branches' (SKILL.md).
- [EXTERNAL_DOWNLOADS]: The skill instructions (SKILL.md) direct the agent to check live configuration files at '/reflex/llms.txt', '/reflex/agent.md', and '/reflex/agent.json' to determine current API/SDK availability. These references point to resources within the vendor's own environment to facilitate the implementation of the reflex.
Audit Metadata