spec-grounded-design

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

• Third-party content exposure detected (high risk: 0.85). The skill’s Step 1–2 runtime path explicitly fetches “the actual spec” and then runs live probes (e.g., curl to real endpoints) and captures the resulting response/error bodies; those are outsider-authored free text from a third-party API/server that the agent ingests into the LLM context.