usage-view

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The agent is instructed (agent-prompt-template.txt, Tool Call 1) to WebFetch the public pricing page at https://platform.claude.com/docs/en/about-claude/pricing#model-pricing and parse those page values to edit scripts/model-pricing.json and re-run the pipeline, so it fetches and acts on third-party web content as part of the required workflow.

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

• Potentially malicious external URL detected (high risk: 1.00). The agent-prompt template instructs the background agent, on ERROR:UNKNOWN_MODEL, to WebFetch the Anthropic pricing page at https://platform.claude.com/docs/en/about-claude/pricing#model-pricing during runtime and use the page's fields to edit scripts/model-pricing.json and re-run the pipeline, so this external URL is fetched at runtime and its content directly controls the agent's actions.