llm-council / adapters
AGENTS.md0 starsChanged 45 days agoMITView on GitHub
Drop this file (or append its contents) into your project's AGENTS.md, GEMINI.md, CLAUDE.md, .github/copilot-instructions.md, .windsurfrules, or CONVENTIONS.md (Aider). Works with OpenAI Codex CLI, Gemini CLI, GitHub Copilot, Windsurf, Aider, OpenCode, Amp, Cline, Roo, and any agent that reads project instructions. If the full skill folder is available, read llm-council/SKILL.md and llm-council/references/* instead: they're the complete version. When the user says "council this", "run the council", "war room this", "pressure-test this", "stress-test this", "debate this", "quick council", or "deep council", or…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repositoryOne other file it gives its agents.
- llm-councilSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Round 1: independent advisors: If you can spawn sub-agents or tasks, spawn one per seat in parallel, each with a fresh context.
- Very specific: line 20
Multi-model: if a shell and an API key are available and the user wants model diversity, run python llm-council/scripts/council.py …
- Distinctive: line 24
No fabricated facts (label assumptions).
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
LLM Council (agent instructions)
Drop this file (or append its contents) into your project's
AGENTS.md,GEMINI.md,CLAUDE.md,.github/copilot-instructions.md,.windsurfrules, orCONVENTIONS.md(Aider). Works with OpenAI Codex CLI, Gemini CLI, GitHub Copilot, Windsurf, Aider, OpenCode, Amp, Cline, Roo, and any agent that reads project instructions. If the full skill folder is available, readllm-council/SKILL.mdandllm-council/references/*instead: they're the complete version.
When to use
When the user says "council this", "run the council", "war room this", "pressure-test this", "stress-test this", "debate this", "quick council", or "deep council", or brings a real decision with stakes and tradeoffs ("should I X or Y", "which option", "I'm torn between"), run the LLM Council. Don't use it for factual lookups, creation or processing tasks, or trivial choices.
How
- Gather context (≤30s): skim README / AGENTS.md / docs / files the user referenced for the 2-4 facts that make advice specific.
- Frame a neutral brief: options, context and numbers, stakes, unknowns-as-assumptions. If it's too vague, ask one clarifying question.
- Round 1: independent advisors. If you can spawn sub-agents or tasks, spawn one per seat in parallel, each with a fresh context. If you can't, write each advisor sealed (never revise an earlier one after writing a later one).
Seats: Contrarian (hunt the fatal flaw) · First Principles (what are we really solving?) · Expansionist (hidden upside) · Outsider (zero context, flags confusion) · Executor (what happens Monday morning?).
Each returns:
Position · Confidence % · Argument · Weakest assumption · Would change my mind(150-300 words). - Round 2: anonymous peer ranking. Shuffle into Response A-E, rotate the order per reviewer, and have each seat review all of them on specificity, soundness, actionability, and insight. Each ends with
FINAL RANKING: 1. Response ? …plus the biggest blind spot and what everyone missed. Average the ranks into a scoreboard. - Chairman: de-anonymize and write: Bottom line · Confidence (High/Med/Low) · One-way or two-way door · Agrees · Clashes (with the crux) · Blind spots · Recommendation (may overrule the majority) · What would change this verdict (tripwires) · The one thing to do first · Scoreboard.
- Deep mode adds a Domain Expert and a Pre-Mortem seat, a rebuttal round (HOLD/UPDATE), and a red team that attacks the draft verdict before it's finalized.
- Multi-model: if a shell and an API key are available and the user wants model diversity, run
python llm-council/scripts/council.py "<brief>" --models <m1> <m2> <m3>.
Rules
No fabricated facts (label assumptions). No sycophancy. No "it depends" without a direction. Name the crux of each disagreement. Give one first step, not ten. For medical, legal, tax, or safety topics, name the professional to consult.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.