Install the council.
Bring a decision.
Install the shared skill and the native roles for your AI assistant, then ask the council to examine a decision that matters to you.
Use the tools you already work in.
Install the shared skill and the three roles for your AI assistant: advisor, reviewer, and chair. Read its execution notes here.
Codex: use Codex workflows in the ChatGPT desktop app, CLI, or IDE extension. ChatGPT Work: subagent workflows are documented, but this package’s installation, role discovery, and complete council execution there remain unverified. See the OpenAI subagent documentation.
Claude support covers Claude Code and local Code sessions in Claude Desktop. Claude Chat and Cowork are outside this installation scope. Versions and settings can affect execution.
Ask your assistant to set it up.
Paste this into your supported AI assistant.
Read the installation instructions at https://github.com/goolamabbas/portable-llm-council and install the shared skill and the agent definitions for this AI assistant in my personal installation locations. Back up existing copies before replacing anything, then verify discovery. Do not run a council yet.
Your assistant may request permission to change installation files. A new session may be needed before the skill and roles are discoverable.
Prefer manual installation? Follow the maintained instructions ↗ (opens in a new tab)Confirm the skill and roles are discoverable.
Check whether you discover the llm-council skill and the three council roles for this AI assistant. Report the resolved locations and effective model settings if available. Do not run a council yet.
Files being present does not establish that the AI assistant can use them. Confirm discovery before starting a full council. Discovery alone does not verify successful execution.
A good brief gives every perspective something to examine.
Use the llm-council skill to evaluate this decision. Decision: [What must I decide, and by when?] Objective: [What outcome matters most?] Options: [What alternatives am I considering?] Constraints: [Time, budget, people, and commitments.] Evidence: [What do I know, and how do I know it?] Unknowns: [What am I assuming or still need to learn?] Recommend a course of action, explain what evidence would change it, and give me one concrete first step. Distinguish facts from assumptions.
A complete run uses five advisors, five fresh reviewers, and one separate chair. Expect an HTML report and a Markdown transcript. Read the recommendation alongside its assumptions and limitations.
For a filled-in brief, see the workshop-or-course example.
Check your first completed run.
Look for five advisor answers, five fresh reviews, one chair verdict, and both the HTML report and Markdown transcript. Check that the report reflects the returned answers. Incomplete rounds or an inability to preserve separate review contexts must be disclosed.
A full council uses 11 agent runs, so allow more time and tokens than for a direct answer.