runhuman-testing

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

• Third-party content exposure detected (high risk: 0.80). The skill's SKILL.md explicitly tells agents to fetch and follow instructions from external public URLs (e.g., "Fetch the comprehensive agent guide at https://runhuman.com/for_agents_github_actions.md" and to check GitHub releases at https://github.com/volter-ai/runhuman-action/releases) and also accepts arbitrary user-provided target URLs for tests, meaning untrusted third-party content is fetched and its instructions can influence agent actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly tells agents to fetch the CI/CD guide at https://runhuman.com/for_agents_github_actions.md (and to check the GitHub releases page https://github.com/volter-ai/runhuman-action/releases) at runtime so the retrieved content will directly determine the agent's instructions for CI/CD setup, which meets the criteria for an external runtime-controlled prompt source.