| name | orca-model-council |
| description | Run a five-subagent council with independent first opinions, anonymized peer critique, and a final chair synthesis. |
ORCA Model Council
What This Skill Is
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.
Trigger
Use when the user wants:
- serious brainstorming instead of one-shot ideation
- a better product or business idea
- feature evaluation before implementation
- market or wedge evaluation
- a multi-angle written judgment with explicit disagreement
Do Not Trigger
Do not use when:
- the task is already a straightforward implementation job
- one clear research or planning pass is enough
- the extra orchestration would add more ceremony than insight
- the user needs a quick answer, not a council
Council Shape
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.
Default Deliberation Stages
Stage 1: First Opinions
Spin up 5 subagents in parallel when possible.
Each one receives:
- the same core question
- the same constraints
- the same artifact goal
- its own role framing
Each subagent returns:
- initial answer
- strongest supporting reasoning
- biggest uncertainty
- provisional recommendation
Stage 2: Anonymous Peer Review
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:
- which response best matches the actual decision
- which response is overconfident
- which response sees the biggest hidden risk
- what idea from another response should survive into the final synthesis
Stage 3: Chair Synthesis
The parent chair reads:
- all first opinions
- all peer critiques
- the aggregate ranking
- repeated disagreement patterns
Then produces:
- final synthesis
- direct recommendation
- main reasons
- strongest objections
- confidence level
- next action
Output Contract
Use:
templates/council-brief.md
templates/council-opinion.md
templates/council-synthesis.md
The final output must preserve disagreement instead of flattening it into fake consensus.
Quality Bar
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.
Common Failure Modes
- five workers saying the same thing in different words
- using role labels that do not change reasoning
- skipping anonymous critique and jumping straight to a summary
- hiding disagreement in a bland final synthesis
- treating average opinion as automatically correct
Relationship To Other ORCA Framework Skills And Commands
Works with orca-agent-orchestration, orca-business-ideation, orca-research, orca-plan-idea, and orca-evaluate-idea.