brida-reflex-policy-change-impact
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions specify fetching live metadata and availability information from the vendor's resources to verify integration capabilities.
- Evidence: SKILL.md directs the agent to read /reflex/llms.txt, /reflex/agent.md, and /reflex/agent.json.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a surface where untrusted data from the environment is processed to make classification decisions, which introduces a potential vector for indirect prompt injection.
- Ingestion points: The skill ingests policyChange, component, and componentResponsibilities fields as defined in the references/custom-reflex.json schema and environment state described in SKILL.md.
- Boundary markers: The references/playbook.md defines a structured workflow fit, but the skill lacks explicit markers or instructions to disregard potential commands embedded within these natural language fields.
- Capability inventory: The agent is instructed to perform integration and verification tasks using MCP, SDK, or REST interfaces, which may involve system interaction and network operations.
- Sanitization: No evidence of input sanitization or validation logic for the processed fields is provided in the skill files.
Audit Metadata