Coco: The Universal AI Agent Orchestration Framework

Transform your AI coding assistant into a high-performance engineering team, backed by an advisory board of 389 world-class minds.

Free | MIT License | 90-Second Onboarding

     Install        Superintelligence        Core Architecture        Skills        Systems        Demos  

git clone https://github.com/rkz91/coco.git && cd coco && bash install.sh

Coco's signature differentiator is Superintelligence. It allows you to summon custom, real-world expert panels directly inside your coding sessions to guide architectural, engineering, risk, finance, and business decisions:

/SI-Decide "Should we migrate our database to pgvector?"
/SI-AI-Decide "Which embedding model gives the best cost/performance?"
/SI-Eng-Pre-Mortem "Review our zero-downtime cache migration strategy"
/SI-GRC-Review "Check this customer onboarding flow for GDPR compliance"

• 389 Expert Personas: Organized across 9 specialized departments:
  • Engineering (70 experts): Experts in software architecture, cloud systems, systems coding, compilers, and infrastructure.
  • AI (59 experts): Neural network research, LLM optimization, safety, alignment, and vector database engineering.
  • Product and Design (56 experts): Design systems, UX design, product strategy, and growth loops.
  • Finance (47 experts): Valuation, corporate finance, macroeconomic modeling, and fintech infrastructure.
  • Trading (46 experts): Quantitative analysis, market microstructure, derivatives pricing, and crypto liquidity.
  • Risk and Compliance / GRC (30 experts): GDPR, HIPAA, SOC2 compliance, security auditing, and international regulations.
  • Strategy (29 experts): Platform economics, business models, competitive analysis, and strategic growth.
  • Data and Analytics (29 experts): Data engineering, pipeline optimization, predictive modeling, and analytics stacks.
  • Sales / GTM / Marketing (23 experts): Product-market fit, sales operations, growth marketing, and enterprise GTM.
• 242 Generated Slash Commands: Comprises 17 Cross-Team Meta Commands (like /SI-Decide or /SI-Tradeoff routing across multiple domains) and 225 Per-Team Commands (25 per team, such as /SI-AI-Decide or /SI-Eng-Pre-Mortem).
• Cross-Team Meta-Orchestration: For complex queries spanning domains ("build and sell an AI audit tool"), a Stage-A local router uses nomic-embed to score the prompt against team registries and cells, greedily picking a proportional 16-32 person panel.
• Citable and Grounded: Every stance is verified from public sources and carries a direct evidence URL. Personas are validated against a strict anti-fabrication gate.

The Meta-Orchestrator uses a staged algorithm to avoid loading massive file systems at runtime — two routing stages (team/cell, then scoped personas) followed by an approval-and-execution step:

1. Stage A: Team and Cell Routing (Cheap): Reads the meta-registry (superintelligence/registry.json) and scans only the team descriptions and cell definitions. It scores team relevance using keyword and domain overlap, selecting the top 1-4 teams. If a single team dominates, it delegates to that team's normal orchestrator.
2. Stage B: Scoped Persona Selection (Scoped): Loads compact persona records (slugs, cells, domains, stances, conflict mappings) for only the selected teams. It allocates the 16-32 panel budget proportionally to team relevance, runs the greedy scorer (0.40·domain + 0.30·cell-coverage + 0.30·conflict-pairing), and applies a cross-team tension pass to pair opposing viewpoints (e.g., security vs. growth).
3. Stage C: Approval and Execution: Presents the roster grouped by team with a one-line selection rationale. The executing action verb consumes the approved roster, and the resulting output includes full line-by-line attribution (e.g., "Aswath Damodaran (Finance): ...").

The 389-persona roster was compiled using a systematic multi-tier workflow:

• Generation: Evaluated Local Qwen 3.6 (via LM Studio) -> GPT-5-Nano (via QB Gateway) -> Gemini 3.5 Flash -> Claude Research Agents (web search). Claude research agents proved to be the quality winner for resolving historical data and cited Signal.
• Validation Gate: Run via validate_persona.py, enforcing that every persona must have at least 4 live, non-404 cited evidence URLs, a valid home team, no fabricated quotes, and at least 2 verified signals within the last 12 months (or archetype persistent signals for historical figures).

A standard installation equips your workspace with a lightweight set of tools, while full activation unlocks up to 864 total assets to orchestrate any software engineering workflow:

Coco comes loaded with 146 skills. These are not basic prompt templates, but fully realized agent instructions containing state management, verification logic, and error handlers:

• ui-ux-pro-max: Generates modern visual presets containing 50+ styling tokens, 21 color palettes, font pairings, and responsive charts.
• axiom-liquid-glass: Implements high-end glassmorphism visual styling specifications using HSL tailored colors, backdrop filters, and subtle border shadows.
• swiftui-liquid-glass: Tailors the visual glassmorphic design system specifically for Apple SwiftUI layouts.
• web-design-guidelines: Enforces strict layout grids, typography scales, responsive query setups, and color harmony constraints.
• tailwind-patterns: A library of production-ready, accessible Tailwind CSS layout patterns (navigation, tables, carousels, cards).

• c4-architecture: Guides the creation of system architecture diagrams following the C4 model (Context, Container, Component, Code).
• api-design-principles: Enforces RESTful and gRPC API standards, focusing on clean resource naming, status code mappings, and error shapes.
• arb-review: Prepares codebases and designs for Architecture Review Board audits, assessing scalability, security, and portability.
• vercel-react-best-practices: Enforces Next.js App Router rules, React Server Components boundaries, and optimized hook placement.
• subagent-driven-development: Instructs the primary AI on how to delegate complex engineering tasks to sandboxed, specialized subagents.

• xlsx: Creates and formats data-dense Excel spreadsheets locally using Python openpyxl scripts.
• pdf: Generates structured PDFs (reports, specs, invoices) with custom tables and headers.
• docx: Assembles word documents with standard paragraph, heading, and table styling.
• doc-sync: Automates synchronization between code implementations and project documentation (PRDs, readmes, APIs).

• dr-plan: Generates disaster recovery specifications containing RTO/RPO targets, failover steps, and data integrity verification.
• recovery-plan: Formulates active incident response plans to restore services after database corruption or server failure.
• using-git-worktrees: Sets up parallel, isolated development branches on the local disk using git worktrees.
• finishing-a-development-branch: Automates the PR prep process: cleans local branch, squashes commits, runs lint, updates changelogs, and opens PR.

• prd-mastery: Generates deep Product Requirement Documents containing user stories, scope limits, metrics, and risk mitigations.
• stakeholder-comms: Generates clear, non-technical progress summaries and release notes tailored to stakeholders.
• nfr-tracker: Audits and tracks Non-Functional Requirements (latency, cost, throughput, security) against defined SLAs.
• brainstorming: Conducts structured brainstorming sessions (SCAMPER, First Principles, Mind Mapping) and outputs formatted markdown tables.

System bundles are opt-in packages that extend the workspace engine with specialized databases, pipelines, and persona rosters:

An orchestration engine consisting of 68 skills and 24 agents that manages the lifecycle of complex codebases.

• State Database: Uses a disk-backed JSON/YAML phase database that records active goals, decisions, milestones, and blockages.
• Workstreams: Spins up parallel workspaces (isolated git checkouts) to test refactors or debug features in sandbox environments without polluting main.
• Forensics: Performs post-mortem audits when a phase fails. Scans commit histories, state files, and agent logs to generate structured root-cause reports.
• Autonomous Mode: Bypasses confirmation steps, allowing parallel waves of subagents to execute plan milestones end-to-end.

A local knowledge graph engine consisting of 6 skills that connects email, chat, code, and documentation.

• Local SQLite Store: Indexes code entities (classes, components), business decisions, project terms, and stakeholder feedback.
• Wiki Generator: Automatically builds local, hyperlinked wiki articles for every recorded entity in the project.
• Mail Sync: Ingests and links project email threads directly to related code files.

A workspace multi-agent pipeline engine.

• Development Loop: Spawns specialized subagents (Research, Architect, QA) to execute changes, review diffs, and write unit tests.
• Deterministic Gates: Intercepts merges and commits to run lints, import audits, reference checks, and verification suites.

/coco yolo

/clone-website https://stripe.com

/code-verification

/systematic-debugging

/prd-generator

/ui-ux-pro-max

bash install.sh --systems brain
/brain-init && /brain-update

/coco

/media-memory

# Host-native scheduling (e.g. Claude Code) pointed at any Coco command:
/schedule run /team:review every Monday
/loop 5m /team:verify

/email:summary
/email:today

/gsd-workstreams create feature-a
/gsd-new-workspace

bash adapters/codex/install.sh

flowchart LR
  U["You<br/>(Slash command, plain English)"] --> AI["Your AI Assistant<br/>(Claude Code, Cursor, etc.)"]
  AI --> A["IDE Adapter"]
  A --> S["Coco Assets<br/>skills · agents · commands · rules"]
  S -. Read at runtime .-> AI
  AI --> Agents["Parallel Subagents<br/>(Execution waves)"]
  Agents --> AI

git clone https://github.com/rkz91/coco.git
cd coco
bash install.sh

# Authenticate GitHub Packages registry
echo "@rkz91:registry=https://npm.pkg.github.com" >> ~/.npmrc
echo "//npm.pkg.github.com/:_authToken=YOUR_TOKEN" >> ~/.npmrc

# Install and launch
npm install -g @rkz91/coco-cli
coco

# Enable GSD, Brain, and Team systems
bash install.sh --systems gsd,brain,team

# Enable Superintelligence personas
bash install.sh --systems superintelligence

bash install.sh --adapter claude-code
bash install.sh --adapter cursor
bash install.sh --adapter codex
bash install.sh --adapter generic

Coco tells you when a newer version is available — it checks the GitHub repository (the source) at most once per day, prints a one-line banner, and sends no telemetry. Disable with COCO_NO_UPDATE_CHECK=1.

# print version + update status
bash scripts/check-update.sh

# apply an update
git pull --ff-only && bash install.sh

# print version + check for updates
npx @rkz91/coco-cli version

# apply an update
npx @rkz91/coco-cli update

_{Releases are tracked in CHANGELOG.md and on the GitHub Releases page.}

_{Core install ships 63 skills + 35 commands + 10 agents. The counts above reflect a full install with all four system bundles enabled (bash install.sh --systems gsd,brain,team,superintelligence). Superintelligence slash commands are generated locally at install time from the team registries — no command files are transmitted or stored remotely.}

Coco is MIT-licensed and contributions are welcome.

• Start here: CONTRIBUTING.md — how to add a skill, command, agent, or adapter.
• Good first issues: help wanted / good first issue — small, well-scoped tasks for newcomers.
• Report a bug / request a feature: open an issue.
• Security: see SECURITY.md for responsible disclosure.
• Conduct: see CODE_OF_CONDUCT.md.

Before opening a PR, validate locally — the CI runs a frontmatter linter, manifest validation, and an install-script syntax check.

Coco is built on top of excellent open-source frameworks and standards. Special thanks to:

• obra/superpowers (Jesse Vincent) for the foundational systematic debugging and verification skills.
• gsd-build/get-shit-done for the 68-skill GSD project orchestration framework.
• JCodesMore/ai-website-cloner-template for the site cloning structure.
• agents.md community for the vendor-neutral agent context standard.
• nickwinder/synthteam for the multi-agent debate protocol behind --debate deliberation.

Full attributions — including feature-level inspirations and licenses — live in CREDITS.md. If something influenced CoCo and isn't listed, that's a bug: open an issue.