define-workflow-refine

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines an 'External Scan Protocol' that ingests untrusted data from the web using search and browser tools. This creates an attack surface where malicious instructions embedded in third-party websites could attempt to influence the agent's requirements analysis.
  • Ingestion points: Web search results and external document retrieval during the Phase 1.4 External Scan.
  • Boundary markers: The skill requires output to be mapped to a rigid Fact/Pattern/Inference/Unknown/Adopt/Reject template to prevent raw content from being treated as direct requirements.
  • Capability inventory: Access to WebSearch, browser, and file system reading tools (Glob, Grep, Read). The skill also dispatches subagents to process data.
  • Sanitization: No specific content filtering or sanitization is described, relying on the agent's instructions to filter data into structured artifacts.
  • [SAFE]: The skill implements a critical security boundary labeled 'HARD-GATE' which explicitly forbids the agent from writing code, invoking implementation tools, or creating project scaffolding before the user has reviewed and approved the specification document.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 06:47 AM
Security Audit — agent-trust-hub — define-workflow-refine