fix-bug

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

• Third-party content exposure detected (high risk: 0.95). The workflow explicitly uses gh issue view ... / WebFetch to read the full GitHub issue and then extracts bug description, code snippets, and reproduction links from that outsider-authored issue text, which is ingested into the agent’s LLM context at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill explicitly instructs the agent at runtime to fetch external content (e.g., raw GitHub raw URLs like raw.githubusercontent.com and online sandbox URLs such as stackblitz.com or codesandbox.io via WebFetch / gh) and to use that fetched issue/code to drive what the agent does (reproduce the bug, write tests, and implement fixes), so the external content directly controls agent prompts and the behavior of the workflow.