llm-council

llm-council is a skill for Claude Code from contactandrewchl-wq/turtle-mcp. It costs 87 tokens per session (2,083 once invoked), scanned A, original, MIT.

A decision-review method in which five opposing AI viewpoints examine an important choice, critique one another anonymously, and produce a traceable summary. It is intended for decisions that are costly or difficult to reverse.

In plain words
What is it for?
Use it to review architecture, data-model, public-contract, technology-stack, or pricing decisions when a mistaken choice could be expensive.
Why use it?
It helps expose weak assumptions, blind spots, incorrect problem framing, and the tendency to agree too quickly with the first answer.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: mentions CLAUDE.md.

Part of the turtle plugin — 26 skills shipped together

Good fit Use it to review architecture, data-model, public-contract, technology-stack, or pricing decisions when a mistaken choice could be expensive.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/contactandrewchl-wq/turtle-mcp/llm-council
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add contactandrewchl-wq/turtle-mcp --skill llm-council
Clone the repo
git clone --depth 1 https://github.com/contactandrewchl-wq/turtle-mcp

Made for: Claude Code.

Or install turtle, the plugin that ships this one along with the rest of its 26 skills.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for llm-council

README.md
[![agentmods](https://agentmods.dev/badge/skills/contactandrewchl-wq/turtle-mcp/llm-council/github.svg)](https://agentmods.dev/skills/contactandrewchl-wq/turtle-mcp/llm-council)
Your own site
<a href="https://agentmods.dev/skills/contactandrewchl-wq/turtle-mcp/llm-council"><img src="https://agentmods.dev/badge/skills/contactandrewchl-wq/turtle-mcp/llm-council/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for llm-council

Your own site · 80×15
<a href="https://agentmods.dev/skills/contactandrewchl-wq/turtle-mcp/llm-council"><img src="https://agentmods.dev/badge/skills/contactandrewchl-wq/turtle-mcp/llm-council.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 87 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,083 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00087 $0.02083
Opus 5 $0.00044 $0.01042
Sonnet 5 $0.00017 $0.00417
Haiku 4.5 $0.00009 $0.00208

Measured 11d ago against content hash 005f3bdf7359, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

llm-council scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/llm-council/SKILL.md · 103 lines

How it starts

The opening of the file, as written. The whole thing — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Consejo deliberativo (LLM Council, sobre Turtle)

Forzar el desacuerdo productivo antes de decidir. En vez de confiar en la primera respuesta (cómoda y complaciente por defecto), la decisión se somete a cinco voces con perspectivas enfrentadas que la atacan desde ángulos distintos, se revisan entre sí en anónimo, y un presidente sintetiza un veredicto con su próximo paso. El objetivo no es consenso tibio: es exponer el modo de falla, el encuadre equivocado y el punto ciego antes de que cuesten caro.

Cuándo usar

Cuando una decisión cumple al menos una:

  • Pesa y es difícil de revertir (arquitectura, modelo de datos, contrato público, elección de stack, pricing).
  • Tenés una respuesta y te convence demasiado rápido — señal de juicio rápido, no de análisis.
  • El encuadre del problema es dudoso: quizá estás resolviendo la pregunta equivocada.
  • Hay sesgo de complacencia: el agente tiende a darte la razón en lugar de discutir.

Si la decisión es trivial, reversible de un commit o ya tiene un camino obvio y barato, no convoques al consejo: decidí directo (ver [[ponytail]]). El consejo cuesta deliberación; usalo donde el error es caro.

Límite de alcance (leer primero)

Este método no spawnea, lanza ni controla procesos, en línea con el SRS §1.2 y con [[agent-orchestration]]:

"TURTLE no es un orquestador que lance o controle agentes; la comunicación entre agentes es asíncrona y mediada por la base de datos."

Las "cinco voces" no son cinco procesos. Son cinco perspectivas. Se materializan de una de dos formas, nunca spawneando:

  • Modo mono-sesión (por defecto): una sola sesión recorre las cinco voces como lentes secuenciales (estilo "sombreros de pensamiento"). Barato, sin coordinación externa, sirve para la mayoría de las decisiones.
  • Modo por el bus (cuando hay personas activas): el presidente convoca por difusión (message_send) a personas reales del roster que aportan su voz por la bandeja, de forma asíncrona. Útil cuando la decisión cruza dominios reales (ej. que [[security-owasp]] hable por boca de Raphael). Sigue siendo bus asíncrono: nadie lanza a nadie.

Read the full file on GitHub · 103 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 11d ago First seen · 103 lines · 87 tokens per session scan A 005f3bdf7359

Subscribe to this mod's changes

llm-council is a skill published in the GitHub repository contactandrewchl-wq/turtle-mcp (2 stars, last pushed 2mo ago), licensed MIT. It adds 87 tokens to every session and 2,083 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

gno

Search local documents, files, notes, and knowledge bases. Index directories, search with BM25/vector/hybrid, get AI answers with citations. Use when user wants to search files, find documents, query notes, look up information in local folders, index a directory, set up document search, build a knowledge base, needs…

gmickel/gno · 86 tokens

engram-memory

Give the agent durable, local memory with engram — recall past decisions before answering, and persist new decisions, preferences and facts as they happen. Use when work spans sessions or the user says "remember".

jnMetaCode/skillet · 45 tokens

simplicio-loop

Unified public entrypoint for Simplicio's body-of-work orchestration: core + loop in one command. Use when the user types /simplicio-loop, says "ralph loop", "keep iterating until done", "finish all open issues", or asks to drain a queue of work autonomously. Runtime-agnostic: binds a real stop-hook where the host…

wesleysimplicio/simplicio · 120 tokens

simplicio-nest-gate

N-Nest protocol — multi-level gate-corrected agent tree. PRIME depth verification, BH port.port.port addressing, per-level confabulation catch.

wesleysimplicio/simplicio · 39 tokens

simplicio-orient

Terminal-first execution — answer facts with the shell, never with the LLM. Use whenever a step needs a fact about the filesystem, git, processes, or system resources, or runs a build/test/lint/diff whose output would flood context. Substitutes deterministic shell/CLI calls for native LLM operations and clamps their…

wesleysimplicio/simplicio · 119 tokens

simplicio-autoresearch

Evolutionary optimize-by-metric loop — mutate a target, evaluate against fixed criteria, KEEP if the score improves (commit) or REVERT if it doesn't (git checkout), repeat, plateau-break after N stagnated runs. Adapts Karpathy's autoresearch pattern (and the ECC autoresearch-agent) with mandatory yool guardrails (§11…

wesleysimplicio/simplicio · 0 tokens