LLM Council

Trust Score 94/100

Run decisions through a council of five AI advisors who independently analyze, peer-review, and synthesize a clear recommendation (Karpathy-style).

triggers:council thisrun the councilpressure-test thisstress-test thisdebate thisshould Iwhich option

GitHub SKILL.md

What it does

Implements a structured LLM Council decision workflow: frame a question with workspace context, run five distinct advisors in parallel (Contrarian, First Principles, Expansionist, Outsider, Executor), run anonymized peer reviews, and synthesize a chairman verdict with clear recommendation and a single next step. Produces an HTML report and full transcript for records.

When to use it

Use this skill for high-stakes decisions or trade-offs where multiple perspectives materially improve outcomes — product strategy, pricing, positioning, launches, and critical business choices. Not for trivial factual queries or simple content generation.

What's included

Scripts: none bundled (has_scripts=false).
References: detailed sub-agent prompts, report/template guidance, and examples embedded in the skill body.
Instructions: full orchestration steps (framing, parallel advisors, peer review, chairman synthesis, report + transcript saving) and precise output formats.

Compatible agents

Best used by multi-agent or subagent-capable systems (Claude Code, ACP/agent runtimes) that can spawn parallel sub-agents and produce HTML reports. Works with agents that can read workspace files for context enrichment.

Audit Summary

LLM Council is a decision-making methodology skill based on Karpathy's multi-model deliberation approach, adapted for single-model sub-agents. It spawns 5 advisors (Contrarian, First Principles, Expansionist, Outsider, Executor) for independent analysis, then peer review, then chairman synthesis. No scripts bundled — purely instructional SKILL.md with detailed templates and examples. Clean, no security concerns.

Watch Out

No scripts to audit — pure instructional skill, static analysis only
Requires sub-agent spawning capability to function
Can be token-intensive (5 advisors + 5 reviewers + 1 chairman = 11 sub-agent calls)

Notes

Well-crafted skill with clear triggers, step-by-step workflow, prompt templates, and output format. Based on Karpathy's LLM Council methodology. No security issues whatsoever — purely text generation with local file output.

Information

Repository: llm-council-skill
Stars: 208

Trust Score

Overall94

Security98

Code Quality92

Architecture88

Usefulness82

Related Skills

Maestro Orchestrate

Orchestrates complex engineering workflows with enforced design and approval gates, child-agent delegation, and session state management.

Run Execute

Orchestrates execution of work items across modes (autopilot, confirm, validate) with scripted init/complete tooling, plan/test/report artifacts and strict gati

Execution Plan Splitter

Break large goals into 30/60/90-day execution plans with milestones, resource needs, and stop/go criteria for practical roadmaps.

Ip2Location IO Automation (Composio)

Automate Ip2location IO tasks via Composio's Rube MCP toolkit — discover tools, verify connections, and execute schema-compliant workflows safely.

BK-CI Design Patterns Guide

Practical guide to design patterns used in the BK-CI project (Kotlin/Spring). Covers factory, strategy, singleton, decorator, observer, template method, chain-o

PPT as Code

Creator-first workflow to plan and build HTML-first presentations with staged modes (quick/basic/advanced), optional PPTX export, and clear QA and image plannin

Build Unplanned Feature

Orchestrates ad-hoc feature development: requirement gathering, planning, implementation and validation across task-planning, implementation, and validation loo

Absolute Prune

Dead code and dependency cleanup across the entire repository.

Back to Skills