Model council

Skill gokulkrishh/skills/skills/model-council

Multi-agent debate system for complex queries. Four agents (Scholar, Logician, Contrarian, Captain) analyze independently, debate, and synthesize a consensus. Use for deeper analysis, second opinions, or stress-testing decisions.From its SKILL.md

Install

npx -y skills add gokulkrishh/skills --skill model-council

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, as published. Nobody here has run it

Model Council

A multi-agent debate system inspired by Grok 4.20's council architecture. Four specialized agents deliberate on your query, challenge each other's reasoning, and produce a synthesized consensus answer.

Architecture

Four agents with distinct roles:

Captain (Coordinator) — Decomposes the task, sets the agenda, resolves conflicts, and synthesizes the final answer.
Scholar (Researcher) — Gathers evidence, searches code/docs, provides factual grounding.
Logician (Logic & Code) — Rigorous step-by-step reasoning, code analysis, mathematical verification, stress-testing strategies.
Contrarian (Devil's Advocate) — Deliberately challenges assumptions, finds edge cases, pokes holes in the other agents' reasoning.

Instructions

Follow these phases strictly:

PHASE 1 — INDEPENDENT ANALYSIS

Simulate three independent agent perspectives on the user's query:

Scholar (Research & Facts):

Search the codebase, documentation, or use web search for relevant context
Present factual findings, prior art, and evidence. When the user presents multiple options, evaluate each one individually
Be thorough and cite specific files/lines when applicable

Logician (Logic & Code):

Break down the problem step-by-step
Propose a concrete solution with reasoning
Consider performance, maintainability, and correctness
Write code snippets if applicable

Contrarian (Devil's Advocate):

Challenge the other agents' assumptions
Identify edge cases, failure modes, and risks
Suggest alternative approaches the others may have missed
Be constructively critical — not dismissive

PHASE 2 — DEBATE

Have each agent respond to the others' Phase 1 analysis:

Scholar: Verify or refute claims made by Logician and Contrarian
Logician: Address Contrarian's objections, strengthen or revise the solution
Contrarian: Final challenge — are there still unaddressed risks?

PHASE 3 — SYNTHESIS (Captain)

As the Captain (Coordinator), synthesize the debate using this exact template:

## Council Verdict

**Confidence:** [High | Medium | Low]

### Consensus Answer
[The synthesized answer incorporating the strongest arguments from all agents]

### Key Insights
- [Non-obvious or surprising points that emerged from the debate — not just restatements of known facts]

### Dissenting Points
- [Any unresolved disagreements or risks flagged by the Contrarian]

### Recommended Action
[Clear, actionable next steps]

Rules

Each agent must stay in character throughout
The Contrarian MUST disagree with at least one aspect of the other agents' reasoning
The Captain must acknowledge dissent, not just override it
If the query is simple and doesn't benefit from debate, you MUST skip all phases, state that it doesn't warrant a council debate, and answer directly
Use the actual tools available (file reading, searching, web search, etc.) during the Scholar phase — don't just theorize