一键导入
orca-model-council
Run a five-subagent council with independent first opinions, anonymized peer critique, and a final chair synthesis.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Run a five-subagent council with independent first opinions, anonymized peer critique, and a final chair synthesis.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Create a host-native `/goal` prompt for bounded ORCA plan work, especially when the user wants an agent to keep working through ready ORCA plans, delegate blocker fixes to subagents, record true blockers, and move on without stalling.
Implement planned changes with scoped edits, verification, and respect for existing project patterns.
Create bounded delegation briefs and structured executor returns for ORCA Framework workflows.
Prepare release readiness checks, release notes, rollback guidance, and handoff.
Record time, retries, and optional token, cache, and cost metrics for ORCA Framework workflows without fabricating precision.
Provide lightweight, skill-sensitive coaching without slowing down normal ORCA Framework execution.
| name | orca-model-council |
| description | Run a five-subagent council with independent first opinions, anonymized peer critique, and a final chair synthesis. |
A reusable ORCA Framework deliberation pattern adapted from the LLM Council idea: multiple subagents answer independently, review each other anonymously, and a parent chair synthesizes the final judgment.
This is for hard judgment calls, not routine execution.
Use when the user wants:
Do not use when:
Default council size: 5 subagents plus 1 parent chair.
The parent talks to the user. The five workers do not.
Use role diversity, not model-brand theater. When native subagents are unavailable, preserve the same council structure as sequential pseudo-subagents with isolated context packets.
Spin up 5 subagents in parallel when possible. Each one receives:
Each subagent returns:
Anonymize the stage-1 outputs as Response A-E.
Each subagent reviews the other responses, not its own. Require each reviewer to rank the responses and explain:
The parent chair reads:
Then produces:
Use:
templates/council-brief.mdtemplates/council-opinion.mdtemplates/council-synthesis.mdThe final output must preserve disagreement instead of flattening it into fake consensus.
The council should produce a better decision, not just more words. A good council output makes tradeoffs, uncertainty, and the next action clearer than a single-agent answer would.
Works with orca-agent-orchestration, orca-business-ideation, orca-research, orca-plan-idea, and orca-evaluate-idea.