publish-models

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

• Third-party content exposure detected (high risk: 0.80). The skill's required cog-safe-push workflow (cog-safe-push.yaml schema) explicitly accepts external URLs in test cases (e.g., match_url: "https://example.com/reference-cat.png" and train.test_cases input_images: "https://.../training.zip") which the tool will fetch/compare as part of validation, so untrusted public content can be ingested and influence whether a push proceeds.

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

• Potentially malicious external URL detected (high risk: 0.90). The CI/workflow fetches and executes remote repository code at runtime — e.g., pip install git+https://github.com/replicate/cog-safe-push.git and the reusable workflow replicate/model-ci-template/.github/workflows/template.yaml@main are installed/used during runs and are required for the skill.