fix-issue

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

• Third-party content exposure detected (high risk: 0.85). The required workflow explicitly uses gh issue view <number> --repo remix-run/remix (or WebFetch) to read the full GitHub issue text at runtime, which is outsider-authored free-form content (issue body/comments) and can include indirect prompt-injection payloads.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill instructs the agent to use WebFetch and raw GitHub/sandbox URLs at runtime (e.g., https://raw.githubusercontent.com/, https://stackblitz.com/, https://codesandbox.io/) to fetch external code/examples that are injected into the agent’s context to reproduce and drive the fix, so those remote URLs directly control prompts/behavior.