knowledge-plan

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

• Third-party content exposure detected (high risk: 0.75). Potential outsider free-text enters via optional Step 0.4 “Preliminary scoping scan” where Phase 0.4 runs 2–3 WebSearch queries, and the resulting web snippets/observations are fed into the agent’s working context for later prompting (“they shape the options you surface and carry them into Step 2 decomposition”).