test-guide-from-code

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

• Third-party content exposure detected (high risk: 0.75). Runtime path Step 1 “Obtain Changes” can fetch MR/PR diffs from GitLab/GitHub via CLI (e.g., glab mr diff, gh pr diff) or prompt the user to paste diff; those diffs are outsider-authored free text (other contributors’ MR/PR content) that the agent then parses and uses to generate the LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches MR/PR diffs at runtime from remote URLs (e.g., https://gitlab.example.com/merge_requests/123 or GitHub URLs containing /pull/) using CLI commands like "glab mr diff" / "gh pr diff", and the retrieved diff content is injected into the agent's context to drive generation of the test guide, so the external URL directly controls the agent's prompts.