CoCo Super Intelligence skill
Orchestration layer for Claude Code, Cursor and Codex: a board of 495 expert personas gives a verdict with named dissent, plus hundreds of skills
- 795 ★ on the repository
- Last commit
Broad access
Why this level
- The installer creates symlinks in `~/.claude`, `~/.cursor` and other editor directories and adds a rules block to `CLAUDE.md`
- Hundreds of skills and commands can run commands and work on your project inside your session
- The optional reverse-skill pack carries reverse-engineering and penetration-testing methods
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/coco-super-intelligenceDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The installer creates symlinks in `~/.claude`, `~/.cursor` and other editor directories and adds a rules block to `CLAUDE.md`
- Hundreds of skills and commands can run commands and work on your project inside your session
- The optional reverse-skill pack carries reverse-engineering and penetration-testing methods
This is third-party code. Review the repository files before installing.
What it does
CoCo Super Intelligence turns Claude Code, Cursor or Codex into an engineering department. Its core is a board of 495 personas across 13 departments: engineering, AI, product and design, finance, trading, legal, risk and compliance and more. Each persona is modeled on the public writing of a named expert, answers from cited public stances and does not express that person's views. `/SI-Decide` takes a question in one line. The orchestrator scores personas on domain fit, expertise-cell coverage and productive conflict, seats 16–32 of them and returns one pick with the dissenters named. The `--debate` flag adds reaction rounds before the tally. A stance check flags claims that do not trace to a source, and stances applied beyond their source are marked as extrapolated. The same install carries skills and commands for the work that follows: planning, review, verification, PRDs, pull requests, the GSD phase engine, the Brain knowledge graph, HyperFrames video skills and M0 memory. Phase state is written to disk, so the plan survives `/clear`. Files stay on your machine and there is no telemetry.
Who it is for. For developers and product leads who want a structured take on a hard decision instead of one nameless model opinion.
Good fit when
- A high-stakes call: whether to move to GraphQL, migrate the database to pgvector, which embedding model to pick
- You want a risk review before launch: `/SI-Eng-Pre-Mortem` on a zero-downtime cache migration plan
- You need a compliance check: `/SI-GRC-Review` on a customer onboarding flow for GDPR
- You want to see which experts disagree rather than an averaged answer
- The work spans several sessions and the plan and progress must survive a cleared chat
Not a fit when
- The question is trivial: a 16–32 persona board burns tokens without raising answer value
- You want a specific person's opinion: personas are built from public writing and are not those people's views
- You need a fully open licence: the core is MIT, while Super Intelligence is proprietary and own-use only
- You do not want an installer touching ~/.claude and CLAUDE.md: it creates symlinks and a rules block
Example request
/SI-Decide "Should a 10-person startup move its public REST API to GraphQL this quarter?"Limitations
It follows an open-core model: the core is MIT, while Super Intelligence is proprietary and licensed for own use. You need git, bash, Node.js 14 or newer for npx, and python3, which generates the `/SI-*` commands.
Generation happens on your machine at install time and nothing is uploaded. Most GSD skills need a separate toolkit in `~/.claude/get-shit-done`, which the installer does not yet provide. When the embedding router is offline, persona selection falls back to keywords. The authors cite a sample `/SI-Decide` run on Claude Sonnet 5.5 that cost US$0.41. The reverse-skill pack of 33 reverse-engineering and penetration-testing skills installs only with an explicit flag.
How to disable. Uninstall removes only symlinks whose target is the CoCo clone and runs from the clone root with a `find` over `~/.claude`, `~/.cursor` and other directories; the exact script is in the update and uninstall section. Generated `SI*.md` files and the rules block between `<!-- coco:rules-start -->` and `<!-- coco:rules-end -->` in `CLAUDE.md` must be deleted by hand. `COCO_NO_UPDATE_CHECK=1` turns off the daily version check.
FAQ
How do I install CoCo in Claude Code?
From your home folder run `npx cocosuperintelligence`: it clones the pinned release into `./coco` and runs `install.sh`, which took about three minutes on a fresh machine. The latest main branch installs with `git clone https://github.com/coco-research/coco.git ~/coco && bash ~/coco/install.sh`. Then run `/SI-Decide` in Claude Code.
What does /SI-Decide do?
It takes a question in one line, seats 16–32 of 495 personas and returns one pick with the dissenters named. With `--debate` the personas react to each other in rounds before the tally, so you see objections rather than a false consensus.
Do the personas speak for real experts?
No. Each persona is modeled on the public writing of a named expert and answers from cited public stances, but it is not that person's view or endorsement. docs/personas/TAKEDOWN.md explains how to request removal or correction.
Which editors does CoCo install into?
The installer auto-detects the editor and works with Claude Code, Cursor, Codex and VS Code; narrow it with `--adapter cursor`. Your model and your keys stay yours. The adapter list, including Aider and Amazon Q, is in the repository's adapters folder.
What are the GSD, Brain, HyperFrames, M0 and Cognee bundles?
Six bundles install by default: GSD with phase orchestration, Brain with a local SQLite knowledge graph, Super Intelligence with the board, HyperFrames with video skills, M0 with shared memory in one SQLite file and Cognee with a local graph. `--core-only` skips the bundles and `--systems` picks specific ones.
Discussion
No questions or reviews yet.
Related
SkillsA desktop app where Claude Code, Codex, Gemini and other agents work as a team of specialists in a shared project
An operating layer for Claude Code and Codex: one /do entry point, project memory across sessions, safety hooks, cost telemetry and parallel agents
A skill that turns several Claude Code, Codex and Cursor windows into a team with a coordinator, shared registry and cross-review
A skill with 25 commands to audit and improve the prompts, tools and agent designs of a project in Cursor, Claude Code and Codex
An orchestration plugin for Claude Code, Codex, Gemini CLI and Qwen Code: 39 specialists, parallel subagents, persistent sessions and standalone audits
A skill that convenes a council of thinker personas and runs a structured deliberation on a hard decision, ending with a verdict
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| CoCo Super Intelligence (this entry) | no | 795 | |
| Superpowers | no | 307,845 | development-process skills for clarification, plans, TDD and review, with no expert board |
| Agency Orchestrator | no | 2,420 | assembles roles from a YAML workflow out of one phrase, with no named expert personas |
README in short
The description opens with the problem: a hard call in a session gets one model's opinion with no name and no record of who disagreed. It then shows a real `/SI-Decide` run with its cost and an honest note that the router fell back to keywords. Sections cover the quick start, a table of 13 departments with expert counts, a skill catalog by bundle, six system bundles and the `/team` command with six development stages and seven gates. The release plan, known gaps and update and uninstall instructions follow.