vibe-plan

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

• Third-party content exposure detected (medium risk: 0.30). Skill “vibe-plan” ingests outsider-authored free text by reading full GitHub issues/PRs (including bodies, comments, and diffs) during Stage 0 triage and also during Stage 3 when fetching referenced spec/issue content from hosted trackers.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill and its evals explicitly instruct the agent to fetch and read external issue URLs at runtime (e.g., https://github.com/acme/app/issues/98), so remote issue content can directly control the agent's prompts/context.