25 agents · 4 model tiers · 121 skills · full handoff map
What is junai?
Most AI coding tools are chat assistants — you ask, they answer, and when the session ends, all context is lost.
junai turns GitHub Copilot into a full software delivery pipeline. 25 specialist AI agents — each scoped to a single role like Architect, Implementer, Tester, or Code Reviewer — collaborate through a deterministic state machine that persists across sessions. Every stage transition is logged in a plain-text pipeline-state.json that lives in your repo.
Idea → PRD → Architecture → Plan → Implement → Test → Review → Done
↑ every stage has a dedicated agent, every transition is tracked, every gate is explicit
How junai compares
| Feature |
junai |
Generic AI Chat |
Other Agent Tools |
| State persists across sessions |
Yes |
No |
No |
| Deterministic routing (state machine) |
Yes |
No |
No |
| 25 role-scoped specialist agents |
Yes |
No |
4–6 generic |
| Full SDLC pipeline |
Yes |
No |
Partial |
| Three modes: supervised / assisted / autopilot |
Yes |
No |
No |
| Autopilot watcher (auto-opens next agent) |
Yes |
No |
No |
| Works with your existing Copilot subscription |
Yes |
No |
Needs separate API keys |
Portable — lives in .github/, travels with your repo |
Yes |
No |
No |
| Auditable — all state in git-committed JSON |
Yes |
No |
No |
| 9 MCP tools callable from chat |
Yes |
No |
Varies |
Quick Start
Three steps to get going:
- Install — Search
junai in the VS Code Extensions panel, or install from the Marketplace
- Initialize — Press
Ctrl+Shift+P → junai: Initialize Agent Pipeline. The full agent pool installs into .github/ and the MCP server is auto-configured
- Start building — Open Copilot Chat, select Orchestrator from the agent picker, and say:
New feature: <describe what you want to build>
I want to run this in autopilot mode.
The pipeline handles routing from there.
Prerequisites
| Requirement |
Why |
| VS Code 1.101+ |
Agent mode support |
| GitHub Copilot subscription |
Agents run as Copilot chat participants |
uv on PATH |
Runs the MCP server — install uv (one command, ~30 seconds) |
Three Pipeline Modes
| Mode |
How it works |
Best for |
| Supervised |
Every gate requires your explicit click before advancing |
Learning the pipeline, high-stakes changes |
| Assisted |
Agents route automatically — you only approve key gates |
Day-to-day feature work |
| Autopilot |
All gates auto-satisfied after intent sign-off; the extension watches pipeline-state.json and opens the next agent automatically — zero clicks |
Trusted, well-scoped work |
Switch anytime from Copilot chat:
"Switch pipeline to autopilot mode"
How autopilot works
In autopilot mode, junai watches pipeline-state.json in real-time. When a stage completes:
- Reads the routing decision from the state file
- Opens the correct specialist agent in Copilot chat automatically
- Sends the handoff prompt — the agent starts working immediately
You approve the intent once. The pipeline does the rest.
25 Specialist Agents
Each agent is a deeply crafted instruction file in .github/agents/ — scoped to a single responsibility, model-matched for the task, and wired with handoffs to the next stage.
Model assignments
| Model |
Agents |
| Claude Opus 4.6 |
Anchor, Architect |
| Claude Sonnet 4.6 |
Orchestrator, Planner, PRD, Code Reviewer, Debug, Security Analyst, Prompt Engineer, Mentor, Knowledge Transfer, Project Manager, UX Designer, UI/UX Designer, Accessibility |
| GPT-5.3-Codex |
Implement, Frontend Developer, Streamlit Developer, Data Engineer, DevOps, SQL Expert, Tester, Janitor |
| Gemini 3.1 Pro |
Mermaid Diagram Specialist, SVG Diagram |
All agents share a handoff protocol — each completion writes artefact paths and routing context into pipeline-state.json, so the Orchestrator can cold-start a new session from state alone.
121 Reusable Skills
Skills are modular knowledge packs that agents load on demand — covering everything from testing strategies to design systems. Organized across 10 categories:
| Category |
Skills |
Examples |
| Coding |
20 |
API design, refactoring, code patterns |
| Frontend |
27 |
CSS architecture, design systems, word clouds, brand design, UI styling |
| Workflow |
16 |
Git, CI/CD, deployment workflows |
| Productivity |
11 |
Documentation, planning, automation |
| Media |
10 |
SVG, image processing, visualization |
| Docs |
9 |
Technical writing, README generation |
| Cloud |
6 |
AWS, Azure, infrastructure |
| Data |
6 |
ETL, data pipelines, analytics |
| DevOps |
6 |
Docker, monitoring, infrastructure |
| Testing |
6 |
Unit, integration, E2E testing |
Agents automatically load the right skills based on the task at hand. You can also reference skills directly in chat.
The MCP server provides pipeline operations callable directly from Copilot chat:
| Tool |
What it does |
pipeline_init |
Start a new pipeline (active-pipeline guard built-in) |
pipeline_reset |
Force-clear and restart (bypasses guard) |
notify_orchestrator |
Record stage completion + trigger routing |
set_pipeline_mode |
Switch supervised / assisted / autopilot |
satisfy_gate |
Manually satisfy a supervision gate |
skip_stage |
Skip the current stage (blocked on implement, anchor, tester) |
get_pipeline_status |
Read current stage, mode, and routing decision |
validate_deferred_paths |
Verify deferred artefact file paths exist |
run_command |
Execute CLI commands from chat context |
What Gets Installed
One command. Everything lands in your .github/ folder and travels with your repo.
| Folder |
What's inside |
agents/ |
25 agent definition files — one per specialist |
skills/ |
121 reusable skill modules across 10 categories |
prompts/ |
30 workflow-level prompt templates (ADR, commit, handoff, etc.) |
instructions/ |
24 coding convention files for Copilot context (Python, SQL, FastAPI, Docker, security, etc.) |
plans/ |
Plan templates and backlog scaffold |
agent-docs/ |
Artefact hub, architecture docs, schema references |
handoffs/ |
Cross-session handoff protocol |
tools/ |
MCP server (auto-registered via uv run, no pip install needed) |
Plus at the root level:
| File |
Purpose |
pipeline-state.json |
Live pipeline state — stage, mode, gates, routing, artefacts |
copilot-instructions.md |
Your project context file — junai manages a small <!-- junai:start --> … <!-- junai:end --> section; everything else is yours and never touched |
.vscode/mcp.json |
MCP server registration (auto-configured) |
Your Files Are Safe
junai uses sentinel-delimited managed sections — the same approach used by SSH config and Terraform. Your content is never overwritten.
| File |
On Initialize |
On Update |
On Remove |
copilot-instructions.md |
Created with a small junai section (or appended if yours already exists) |
Only the <!-- junai:start --> … <!-- junai:end --> block is refreshed — your content is untouched |
Only the junai section is stripped — your content stays |
pipeline-state.json |
Created if missing |
Never touched |
Deleted |
project-config.md |
Created (backup if overwriting) |
Never touched |
Deleted |
| Agent / skill / instruction files |
Installed |
Updated to latest |
Deleted |
Commands
| Command |
What it does |
junai: Initialize Agent Pipeline |
Install the full agent pool and configure the MCP server |
junai: Update Agent Pool |
Pull latest agent/skill files — preserves your pipeline state |
junai: Show Pipeline Status |
View current stage, mode, gate states, and last routing decision |
junai: Set Pipeline Mode |
Switch pipeline mode without re-initializing |
junai: Remove from this project |
Clean uninstall — removes agent pool and MCP config |
Extension Settings
| Setting |
Default |
Description |
junai.defaultMode |
supervised |
Pipeline mode applied on Initialize |
Learn More
The future of software engineering is agentic.
junai makes it structured, auditable, and yours.
MIT © junai Labs
| |
