You manage delivery. AI agents ship story by story — with a plan that survives the next chat session.
Meridian is a repo-native harness for AI-assisted development in VS Code and Cursor. It gives your agents structured workflows (slash commands, roles, gates) and gives you a board, planning views, and dependency graphs — without leaving the IDE.
Install: Visual Studio Marketplace — search Meridian Harness, publisher colabcolibri
GitHub: https://github.com/colabcolibri/meridian
Who this is for
You use Cursor, VS Code, Claude Code, or Codex to build software with AI, and you want:
- A real backlog — versions, sprints, epics, and user stories in the repo, not buried in chat history
- Control over “done” — scope and architecture written down before code; evidence when a story closes
- One loop — document → plan → refine → implement → close, repeated story by story
- Agents that follow rules — workflows route the right role; skills and validators keep output consistent
If you only need ad-hoc prompts with no delivery structure, Meridian is probably more than you need.
What you get with this extension
This extension is the recommended way to use Meridian: same harness as the kit, plus visualization inside the IDE.
| You get | Why it matters |
|---|---|
| Kanban board | See backlog, todo, in progress, and done at a glance — filters by version, sprint, epic |
| Planning views | Versions roadmap, sprints, epics, decisions log |
| Delivery & import graphs | Explore story dependencies and how your codebase imports itself |
| Architecture diagrams | Mermaid maps from docs/architecture/ with pan/zoom |
| Bundled kit installer | One click copies .agent/ (workflows, agents, skills) into your project |
| Guides in the sidebar | Onboarding without hunting the repo |
In chat you still run slash commands (/status, /create-us, /implement-us US-0103). The extension is for seeing and validating; chat is for creating and changing the plan.
Harness only (no extension)?
Yes. Meridian works without this plugin.
| With extension (this) | Kit / CLI only | |
|---|---|---|
| Slash workflows in chat | ✅ | ✅ |
.agent/ agents, skills, validators |
✅ (via Install Harness) | ✅ (install script or copy .agent/) |
docs/ + .meridian/meridian.db delivery |
✅ | ✅ |
Python CLI (meridian_delivery.py, export, validate) |
✅ | ✅ |
| Board, graphs, planning UI | ✅ | ❌ — use /status and SQLite export instead |
Use the extension when you want the board and graphs in the IDE. Use the kit alone in headless setups, other editors, or CI — same protocol, no visual layer.
Details: Distribution · Kit README
The problem Meridian solves
Agents write code fast. Without a written plan:
- Scope drifts every session
- Decisions disappear into old threads
- “Done” means whatever the model said last
Meridian keeps what to build (docs/) and what’s in flight (SQLite delivery DB) in the repository. Chat is ephemeral; files and the board are not.
How it works
document → plan → refine → implement → close
- Document — approve scope, architecture, security in
docs/ - Plan — epics, versions, sprints, user stories (
/create-us, …) - Refine — story is concrete enough to build (
/refine-us) - Implement — agent codes only when the story is
ready(/implement-us) - Close — evidence on the story record (
/complete-us); you approve
You type workflows (/create-us), not low-level skills. Workflows route agents automatically. Override with @backlog-refiner when needed.
Quick start
- Install Meridian Harness from the Marketplace → reload the window
- Open your project folder
- Meridian: Install Harness (status bar or command palette)
- In chat:
/init-meridian(new project) or/document-project(existing codebase) - Meridian: Open Board · anytime
/status
Read first: Meridian: How to Use (sidebar → Meridian → Commands → Guides)
Guides in the IDE
| Guide | Purpose |
|---|---|
| How to use | Extension vs chat, first steps |
| Start here | Concepts — phases, gates, artifacts |
| Usage guide | Day-to-day recipes |
| Agents & slash commands | Full command reference |
Main commands
| Goal | Command |
|---|---|
| Onboarding | Meridian: How to Use |
| Kanban | Meridian: Open Board |
| Planning | Open Versions, Open Sprints, Open Epics, Open Decisions |
| Graphs | Open Delivery Graph, Open Import Graph |
| Diagrams | Open Architecture Diagram |
| Health check | Meridian: Validate Project (needs python3) |
| Multi-product monorepo | Meridian: Select Active Project |
Update the harness
- Update Meridian Harness in Extensions
- Reload window
- Meridian: Upgrade Harness in each project — refreshes
.agent/, IDE adapters, and DB migrations
Requirements
- VS Code 1.85+ or Cursor
python3only for Validate Project and CLI scripts (optional for day-to-day chat workflows)
License
PolyForm Noncommercial 1.0.0 — free for noncommercial use. The Meridian kit (.agent/) ships inside this extension.