inbound-triage

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md “Inbound Triage” step 1–2 requires reading the target issue (including its description and comments) and step 2 says to inspect fields such as description and comments, meaning outsider-authored free text placed into those Paperclip issue/comment records will be ingested for triage at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill text references runtime REST/MCP endpoints such as GET /api/agents/{agentId}/instructions-bundle which are used at runtime to fetch managed agent instruction bundles that directly control agent prompts/instructions.