cd /news/ai-agents/council-five-independent-advisors-fo… · home topics ai-agents article
[ARTICLE · art-128344] src=github.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Council – Five independent advisors for your agent

Developer Matt Akins released Council, an open-source skill that runs five independent AI advisors in parallel and has a Chairman subagent synthesize their verdicts into one recommendation. Council works with Claude Code, OpenAI Codex, OpenCode, Cursor, and ChatGPT Work, installs via plugin marketplaces or the command `npx skills add mattakins/council -g`, and supports 3- or 7-advisor runs, a research mode, and a `quick` flag that skips the Chairman. The project is inspired by Andrej Karpathy's LLM Council and targets architectural, product, and naming decisions.

read3 min views7 publishedSep 13, 2026
Council – Five independent advisors for your agent
Image: Michielbdejong (auto-discovered)

Five independent advisors. One clear recommendation.

Council is a skill that helps you make decisions. Ask a question about your project. Five independent advisors weigh in. The Chairman brings it all together. Your agent adds its own take and next step.

Use it for architectural calls, product and marketing decisions, naming debates — any decision worth more than one perspective.

Website · Install · See examples

Works with Codex, Claude Code, OpenCode, Cursor, ChatGPT Work, and other harnesses that support parallel, isolated subagents.

Inspired by Karpathy's LLM Council.

Claude: /council:council should we rewrite this service in Rust?
Codex:  $council:council deep what's the commercial viability of this app right now?
/plugin marketplace add mattakins/council
/plugin install council@mattakins

Plugin skills are namespaced. Invoke this skill as /council:council <question>.

codex plugin marketplace add mattakins/council
codex plugin add council@mattakins

Invoke the plugin skill as $council:council <question>.

npx skills add mattakins/council -g

The Skills CLI installs the shared SKILL.md into the paths expected by the agents you select.

Clone or copy plugins/council/skills/council into one of these supported locations:

  • Claude Code: ~/.claude/skills/council/
  • Codex: ~/.agents/skills/council/
  • OpenCode: ~/.agents/skills/council/ or~/.config/opencode/skills/council/
Command What it does
<council> <question> Default: 5 agents + Chairman
<council> 3 <question> Use 3 agents
<council> 7 <question> Use 7 agents
<council> deep <question> Research + 7 agents + Chairman (full treatment)
<council> research <question> Explicitly enable research
<council> no research <question> Skip research, reasoning only
<council> "SEO, legal, UX" <question> Specify perspectives explicitly
<council> quick <question> Skip Chairman, faster output
<council> no chairman <question> Same as quick
<council> save <question> Write output to council-YYYY-MM-DD-<slug>.md
<council> opus <question> Run all advisors on Opus (any named model works if your harness supports per-subagent overrides)
<council> sonnet <question> Run all advisors on Sonnet

Replace <council> with /council:council for the Claude plugin or $council:council for the Codex plugin. Flags combine, and options also work as plain language — e.g. use 4 advisors, no research, and save: <question>.

  1. Parse — extracts agent count, perspectives, flags, and auto-detects if research is needed
  2. Research detection — if the question involves markets, companies, current events, or real-world data, agents are automatically told to research before answering. No flag needed.
  3. Parallel agents — all advisors run simultaneously, blind to each other, each from their assigned perspective
  4. Chairman synthesis — a fresh, isolated subagent synthesizes the council into a definitive verdict (skippable withquick )
  5. My Take — the final section: an opinionated recommendation drawing on both the council outputs and full conversation context
Chairman's Verdict:
[Definitive synthesis]

| Agent | Perspective | Recommendation | Confidence |
| ...   | ...         | ...            | ...        |

Consensus: ...
Dissent (if any): ...
Risks: ...
Verdict: ...

My Take: ...        ← always present, always opinionated
Next step: ...
Tool Works? Install path
Claude Code Plugin or ~/.claude/skills/council/
OpenAI Codex Plugin or ~/.agents/skills/council/
OpenCode ~/.agents/skills/council/ or native OpenCode path
Cursor Agent Skills install ( npx skills add )
Other harnesses Conditional Requires parallel subagent support

The instruction format follows the open Agent Skills specification. Full behavior requires a harness that can launch parallel, isolated subagents.

MIT

── more in #ai-agents 4 stories · sorted by recency
── more on @council 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/council-five-indepen…] indexed:0 read:3min 2026-09-13 ·