recoup-platform-build-os

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested at runtime from the harness-provided use-input/*.md files (e.g., evals/scenarios/consulting-firm/use-input/acme-followup.md, record-label/use-input/new-release-brief.md, research-lab/use-input/experiment-001-results.md) which the generated {domain}-intake/auto-manage loop reads and then files into knowledge/ and updates dashboards—so that outsider-authored prose becomes LLM context via the agent’s “read it / extract it” steps.