brida-reflex-semantic-response-health

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 classify external transport payloads and project environment state, which introduces a potential indirect prompt injection surface.
  • Ingestion points: Untrusted data enters the agent's context through project environment inspection and response payload summaries defined in the reflex contract (references/custom-reflex.json).
  • Boundary markers: Absent; the instructions do not define delimiters or specific "ignore instructions" warnings to isolate untrusted data from the classification logic.
  • Capability inventory: The skill utilizes tool capabilities for project inspection and integration through MCP, SDKs, or REST APIs for reflex execution (SKILL.md).
  • Sanitization: Absent; no explicit validation, filtering, or escaping of the ingested payload content is described.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to read live configuration and availability metadata from vendor-specific paths such as /reflex/llms.txt and /reflex/agent.json. These resources appear to be vendor-managed artifacts related to the brida-ai infrastructure.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 08:22 PM
Security Audit — agent-trust-hub — brida-reflex-semantic-response-health