agent-evals

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The autonomous controller template (references/templates/autonomous-improve-loop.mjs) performs a runtime POST to https://api.openai.com/v1/responses to generate optimizer output (a unified diff) which the script then uses to produce and apply code patches, so remote content directly controls prompts/instructions and code changes.