jp-architecture

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

• Third-party content exposure detected (medium risk: 0.30). Step 1 ingests outsider-authored PRD/addendum/user-provided notes and possible incoming handoff-<this_side>.md from {project-root}/docs/juspay/ (within step-01-init.md), and subsequent steps analyze that content, so an outsider can submit free text via their repo/docs that the runtime reads before selecting any specific trusted item.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly calls doc_fetch_tool at runtime to fetch Juspay documentation from the allowed external domains (*.juspay.io, *.juspay.in, dth95m2xtyv8v.cloudfront.net), and those fetched docs are required and used to directly ground and control the agent's architecture decisions and outputs.