github-workflows

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs fetching public docs from https://docs.github.com (SKILL.md "Documentation Routing") and — more importantly — its runtime scripts (e.g., references/gh/scripts/github_project_setup.py and experiment_cleanup.py) call the GitHub API / PyGithub (repository.get_issues(), get_labels(), get_milestones(), gh api/graphql) to read user-generated issue/PR/milestone data and then make label/issue/project changes based on that content, so untrusted third-party content is read and can directly influence agent actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly instructs the agent at runtime to fetch live GitHub Actions docs and use their .md content to determine workflow syntax/versions (https://docs.github.com/en/actions/reference/workflows-and-actions/), so that external URL directly controls the agent's instructions and is a required runtime dependency.