test-iterate-loop

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow ingests outsider-authored free text via the project’s failing test logs and then uses Read on the test and implementation files to form hypotheses and fixes, and those file contents can contain arbitrary user-authored text at runtime.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's --mock-hpc mode names a default Docker image "user/avon-mock:latest" (SKILL.md), which would be fetched/pulled and executed at runtime to run tests, so it is a runtime external artifact that executes remote code.