claude-codex-consensus

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow calls an external “other agent” via the agent CLI and requires that agent to write back its exact prompt and findings into a review artifact; that callee’s free-form review text (authored by a different agent/system, i.e., an outsider source) is then ingested into the orchestrator’s LLM context for reconciliation and final reporting.