coco-local

Coco

Autonomous spec-driven development for Claude Code.
Describe a feature. Get merged, tested, reviewed code.

---

Named after Coco, a toy poodle who is small, fiercely opinionated, and relentlessly autonomous -- much like this plugin. She chases birds with the same energy Coco chases tasks through a dependency graph: methodically, loudly, and without asking for permission. When she's not barking at strangers or sneaking cheese, she's napping -- recharging for the next burst of chaotic productivity.

Why Coco?

Most Claude Code workflows handle one slice of the problem. A spec generator here, a task tracker there, an autonomous loop somewhere else. You end up duct-taping three repos together and hoping they agree on file formats.

Coco is the whole pipeline in one plugin:

 Discover        Plan            Decompose       Execute        Review        Ship
───────── ──▶ ─────────── ──▶ ─────────── ──▶ ─────────── ──▶ ──────── ──▶ ──────
 PRD +          interview →     tasks.md +      TDD loop       AI code      PRs merged,
 Roadmap        design.md       tracker import  + commits      review       issues closed

Zero dependencies beyond bash and jq. No daemon. No database. No node_modules. Install from the marketplace and go.

What You Get

• Discovery to delivery -- PRD, roadmap, design, tasks, implementation, review, merge. One plugin.
• Dependency-aware execution -- Built-in tracker with topological sort. ready always returns the next unblocked task.
• Autonomous loop -- Circuit breaker, progress detection, configurable safety limits. Runs until done or stuck. Supports worktree-based parallel execution.
• AI code review -- Every PR reviewed before merge. Critical findings auto-fixed. Up to 3 review iterations.
• Adaptive routing -- Quick fix? Skip the ceremony. Full feature? Full pipeline. Complexity detected automatically.
• Session memory -- Survives context compaction. Pick up where you left off across sessions.
• Multi-repo support -- Derive platform-specific PRDs from a primary repo. Each satellite runs its own independent pipeline.
• Issue tracker sync -- Linear, GitHub Issues, or neither. Config-driven, not hard-coded.

Quick Start

Install

From the marketplace (recommended):

In Claude Code, run:

/plugin marketplace add skullninja/coco-workflow
/plugin install coco@coco-workflow

Then in your project:

/coco:setup

The setup command creates the .coco/ directory, walks you through key settings (project name, issue tracker, parallel execution), installs git hooks, and configures permissions.

As a git submodule (alternative):

git submodule add https://github.com/skullninja/coco-workflow.git coco-workflow
bash coco-workflow/scripts/setup.sh

Build something

# Full pipeline: discovery to delivery
/coco:prd "My product description"       # Create PRD
/coco:roadmap v1.0                       # Prioritized roadmap
/coco:phase "Phase 1: Foundation"        # Orchestrate all features

# Single feature
/coco:planning-session tactical               # Interview -> design -> tasks -> import
/coco:loop                               # Autonomous execution until done

# Quick fix
# Just describe the bug -- hotfix skill handles it

Check in on progress

/coco:standup                            # What happened, what's next, what's stuck
/coco:status                             # Execution state and parallel opportunities

How It Works

Architecture

Layer	What	How
Discovery	PRD, analysis, roadmap, multi-repo derive	/coco:prd, /coco:roadmap
Planning	Discovery, design, task decomposition	AI-selected skills (interview, design, tasks, import)
Execution	Dependency resolution, TDD loop	Built-in tracker + /coco:loop
Review	AI code review on every PR	code-reviewer agent
Visibility	Issue tracker sync	Linear MCP, GitHub CLI, or none

Adaptive Complexity Routing

Not every change needs the full pipeline. /coco:planning-session tactical and /coco:phase automatically right-size the workflow: