Jig

The AI engineering workflow framework for teams.

In manufacturing, a jig holds the workpiece and guides the tool — ensuring every cut is precise, every result consistent. This framework does the same for AI-assisted development.

Quick start

# Add the Jig marketplace (one-time per user)
$ /plugin marketplace add duronext/jig

# Install for your whole team
$ /plugin install jig@duronext-jig --scope project

# Set up your project
$ /jig:init

Three commands. Jig walks you through the rest.


AI made every engineer 10x faster. It also made every mistake 10x faster. Teams with high AI adoption are seeing more bugs, larger PRs, and longer review cycles — not because the AI is bad, but because nothing is holding it accountable.

Without a shared framework, every engineer invents their own workflow. Skills sit in scattered files no one remembers to use. Code review depth varies by who happens to review. A bag of skills is not a workflow — it's vibe coding with extra steps.

The best frameworks don't just hand you tools — they give you opinions. They're the reason a team of two hundred engineers can onboard someone on Monday and watch them ship their first PR by Wednesday. Not because the framework is magic, but because everyone follows the same pipeline, the same quality gates, the same patterns.

Jig brings that discipline to AI engineering. Read the research →


DISCOVER BRAINSTORM PLAN EXECUTE REVIEW SHIP LEARN

Every stage has a skill. Every handoff has a quality gate.

Work type — bug, feature, improvement, task — determines which stages run and how deep they go.


Discover
kickoff
Classifies work, routes through the pipeline
Brainstorm
brainstorm
Design exploration with concerns checklist
prd
Structured requirements with acceptance criteria
Plan
plan
Designs to bite-sized TDD tasks
Execute
build
Picks parallel or serial, runs the plan
team-dev
Spawns parallel agent teammates
sdd
Serial execution with two-stage review
tdd
Red-green-refactor discipline
Review
review
Specialist swarm, parallel reviewers
verify
Evidence before assertions
debug
Root cause before fixes
Ship
pr-create
PR with description and test plan
pr-respond
Analyze, fix, reply to review feedback
finish
Branch completion and cleanup
Learn
postmortem
Extracts lessons, improves skills
Meta
extend
Scaffolds new skills, specialists, packs
ticket
Files issues in your tracker
Agents commit · code-review · pr-review
Specialists security · dead-code · error-handling · async-safety · performance
Starter pack copywriting · logging · testing · test-coverage
DevEx worktree

1
Install

Add one JSON snippet to your project's .claude/settings.json. Every teammate gets the full framework on clone — pipeline skills, review specialists, and agents. Zero manual setup.

2
Configure

Run /jig:init and answer five questions. Jig auto-detects your git host, branch, and commit conventions, then walks you through team name, ticket system, and engineering concerns. It generates your jig.config.md and team directory — ready to go in two minutes.

3
Extend

Add domain skills, custom specialists, team agents. Drop them in the right directory — Jig discovers them automatically. /jig:extend scaffolds new artifacts with valid frontmatter and wires them into the framework.


Jig was born inside Duro, a hardware startup where jigs aren't metaphors — they're tools on the factory floor.

The engineering team spent months building an AI development system for their platform: 31 skills, 6 agents, and 10 specialist reviewers — each one forged from actual use cases, refined through real code review, battle-tested against shipping deadlines.

When they stepped back and looked at what they'd built — a full-lifecycle pipeline that every engineer relied on daily — they realized it shouldn't stay locked inside one company.

So we extracted it, generalized it, and released it as Jig.


A feature, from idea to shipped — guided by Jig at every step.

$ /jig:kickoff
  ▸ Work type: feature
  ▸ Summary:  Add user notification preferences
  ▸ Pipeline: discover → brainstorm → prd → build → review → ship

$ /jig:brainstorm
  ▸ What problem does this solve for users?
  ▸ Approach A: Per-channel toggles with sensible defaults
  ▸ Approach B: Smart digest with frequency control
  ▸ Design approved 

$ /jig:prd
  ▸ PRD: User Notification Preferences
  ▸ 12 acceptance criteria defined
  ▸ Layer tags: [API] [DATA] [UI]
  ▸ Approved 

$ /jig:build
  ▸ Strategy: parallel (4 tasks, 3 agents)
  ▸ Agent 1: Data model + migrations   done
  ▸ Agent 2: API endpoints             done
  ▸ Agent 3: UI components             done
  ▸ Spec compliance: passed ✓
  ▸ Code quality: passed ✓

$ /jig:review
  ▸ Dispatching 7 specialists...
  ▸ security: passed
  ▸ async-safety: passed
  ▸ error-handling: 1 suggestion (minor)
  ▸ Confidence: 94% — ready to ship

Your team's AI workflow, unified.

Add the plugin, run /jig:init to set up, then /jig:kickoff to start your first task.

Built by the Duro engineering team

Forged at Duro · open-sourced for every team