Multi Model Critique

Use multiple models in a 4-step cycle of drafting, cross-critique, revision, and synthesis to generate higher-quality answers for complex, high-stakes queries.

This skill appears to do what it says: orchestrate multiple ACP models through draft → critique → revision → synthesis. Before using it, confirm that you trust the ACP agentIds you will include (the workflow sends user content to all listed models), and ensure the run-time will not point output files to sensitive system locations. Note the templates default to Korean unless overridden. There are no requested credentials or external downloads in the package. If you will run the helper scripts locally, inspect the output directory arguments you provide and ensure the platform's sessions_spawn / sessions_send privileges are granted only to models you trust.

Type: OpenClaw Skill Name: multi-model-critique Version: 1.0.1 This skill bundle implements a multi-model critique workflow with strong, explicit prompt injection mitigations. The Python scripts (`scripts/build_round_prompts.py`, `scripts/run_orchestration.py`) validate and normalize user inputs (`question`, `constraints`), rejecting known prompt-injection control phrases via `DISALLOWED_PATTERNS` and wrapping user content in explicit data blocks with untrusted content warnings. The `SKILL.md` and `references/orchestration-template.md` instruct the OpenClaw agent to use legitimate internal tools (`sessions_spawn`, `sessions_send`, `sessions_history`) for orchestration, without any malicious directives for the agent itself. No evidence of data exfiltration, unauthorized execution, persistence, or obfuscation was found.