bx-ai-pipelines

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill documents a pipeline architecture for processing data through AI models which creates an attack surface for indirect prompt injection. 1. Ingestion points: The skill uses ${variable} placeholders in aiMessage() templates and a ${_input} system variable to pass data between stages in SKILL.md. 2. Boundary markers: Examples show direct interpolation of variables (e.g., Translate '${text}' to ${language}) without using XML tags, delimiters, or 'ignore instructions' warnings to isolate user data. 3. Capability inventory: The skill enables model execution via aiModel() and logic execution via aiTransform() closures. 4. Sanitization: The documentation lacks examples of sanitizing, validating, or escaping external content before it is processed by the AI pipeline.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 09:01 AM
Security Audit — agent-trust-hub — bx-ai-pipelines