recipe-discover

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted data that could influence subsequent file-writing operations.
  • Ingestion points: The skill accepts input via $ARGUMENTS for context assessment, reads results from web search for market research, and incorporates findings from existing code via the codebase-analyzer sub-agent.
  • Boundary markers: Explicit human-in-the-loop checkpoints ([STOP — BLOCKING]) are defined at Step 4 (Opportunity Identification) and Step 5 (Hypothesis Generation), requiring manual user approval before the agent proceeds.
  • Capability inventory: The agent is authorized to write multiple files (Opportunities, Hypotheses, Journeys, and a summary index) to the docs/discovery/ directory upon approval.
  • Sanitization: The workflow relies on manual user verification of drafted findings as a procedural control to prevent external data from injecting malicious content into project files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 10:16 PM
Security Audit — agent-trust-hub — recipe-discover