PolyClaude plugin
Claude Code plugin that convenes a council: members analyze alone, then cross-examine one another and deliver a verdict.
- 186 ★ on the repository
- Last commit
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/polyclaudeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- Parallel subagents spend tokens and API money.
This is third-party code. Review the repository files before installing.
What it does
PolyClaude convenes a council on your question, plan or idea. The members are six built-in lenses (`architect`, `skeptic`, `pragmatist`, `innovator`, `advocate`, `temporal`) or experts you name, such as a typographer and an accessibility specialist. Each first analyzes the question alone, then reads the others' arguments and answers them: challenges, concedes, holds or changes position. The output is not a list of opinions but a verdict: what held under challenge, what moved in cross-examination, which tensions are real trade-offs, which blind spots nobody caught and a confidence map. It runs through `/polyclaude` or `/polyclaude:council`, with a council of two to six members and the flags `--quick`, `--full`, `--deep` for Opus members and `--no-debate`. Members run on Sonnet by default. By the author's estimate a full council costs 2 to 5 dollars at API prices.
Who it is for. Developers and team leads who want an architecture or product decision tested from several sides before choosing, not one confident answer.
Good fit when
- Before rewriting a module, rewrite and incremental patching are compared.
- A migration plan passes through a skeptic and a pragmatist.
- A panel of specialists on a topic outside your field is wanted.
- A quick check of an idea by two council members without debate.
Not a fit when
- A quick answer to a simple question is needed, while a council costs money and minutes.
- The decision is already made and only executors are needed.
- The API call budget is tightly limited.
Example request
Convene a four-member council and decide whether to rewrite the auth system or patch it incrementally.Limitations
The council launches parallel subagents and so spends tokens: two members with no debate cost about 0.7 to 1.1 dollars, six with cross-examination up to 5 dollars at API prices and 12 to 15 minutes. About 80 percent of the spend is the main session. The `--deep` mode on Opus costs roughly 2.5 times more for the members. These are the author's rough figures, and actual cost depends on the brief's length. Claude Code needs a restart after installation.
How to disable. Run `claude plugin uninstall polyclaude@polyclaude` and, if you wish, `claude plugin marketplace remove polyclaude`, then restart Claude Code.
FAQ
How do I set up a panel of my own experts?
Name the roles with `--panel`, for example `--panel "typographer; interface engineer"`, or just in the request. PolyClaude builds a member for each role.
What does --quick do?
A two-member council with one round and no debate: a fast check in about 3 minutes and roughly one dollar by the author's estimate.
Which built-in lenses exist?
`architect`, `skeptic`, `pragmatist`, `innovator`, `advocate` and `temporal`. Exclude unwanted ones with `--exclude` and add wanted ones with `--include`.
Why does installation fail with an SSH error?
The machine is trying SSH for GitHub without keys. Run `git config --global url."https://github.com/".insteadOf "git@github.com:"` and add the marketplace again.
How do I update the plugin?
`claude plugin marketplace update polyclaude`, then `claude plugin update polyclaude@polyclaude` and restart Claude Code.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin that asks several AI models the same question and shows their answers side by side with a synthesis of agreement and divergence
A Claude Code plugin that runs one task across several external models and surfaces disagreements before you ship
Plugins for Claude Code and Codex bring key decisions to the human before coding: /inquire, /elicit, /bound and ten more.
Claude Code supervises while Codex, opencode and pi work on your quota: delegated edits, independent review, model discussions
A Claude Code plugin where the main model plans and signs off while search, edits and verification go to roles running on cheaper models
Orchestration layer for Claude Code, Cursor and Codex: a board of 495 expert personas gives a verdict with named dissent, plus hundreds of skills
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| PolyClaude (this card) | no | 186 | |
| LLM Council Skill | no | 501 | collects answers from several models; PolyClaude builds a council of roles and runs cross-examination within one model |
| claude-council | no | 932 | shows different models' answers side by side; here members read and rebut each other's arguments |
README in short
The result is a verdict with a confidence map and blind spots after cross-examination, not a set of opinions.