https://github.com/divyamohan1993/claude-kanban
Drop an idea. Walk away. Come back to committed code. Autonomous AI pipeline using Claude Code.
https://github.com/divyamohan1993/claude-kanban
ai-automation anthropic autonomous-agents ci-cd claude claude-code code-review developer-tools express kanban nodejs orchestrator pipeline self-healing sqlite
Last synced: 2 months ago
JSON representation
Drop an idea. Walk away. Come back to committed code. Autonomous AI pipeline using Claude Code.
- Host: GitHub
- URL: https://github.com/divyamohan1993/claude-kanban
- Owner: divyamohan1993
- License: apache-2.0
- Created: 2026-03-07T04:58:35.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-04-11T23:12:25.000Z (2 months ago)
- Last Synced: 2026-04-13T19:43:55.380Z (2 months ago)
- Topics: ai-automation, anthropic, autonomous-agents, ci-cd, claude, claude-code, code-review, developer-tools, express, kanban, nodejs, orchestrator, pipeline, self-healing, sqlite
- Language: JavaScript
- Homepage: https://github.com/divyamohan1993/claude-kanban#readme
- Size: 936 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE
Awesome Lists containing this project
README
# ◆ Claude Kanban
### Autonomous AI Build Pipeline
**Drop an idea. Walk away. Come back to committed code.**
[](https://github.com/divyamohan1993/claude-kanban/actions/workflows/ci.yml)
[](https://github.com/divyamohan1993/claude-kanban/actions/workflows/codeql.yml)
[](https://opensource.org/licenses/Apache-2.0)
[]()
[]()
[]()
[]()
```mermaid
flowchart LR
A[Idea] --> B[Brainstorm] --> C[Build] --> D{AI Review\n1-10}
D -- "Score ≥ 8" --> E[Auto-commit\n+ push]
D -- "Score 5-7" --> F[Auto-fix ×3\nthen re-review]
D -- "Score < 5" --> G[Human review]
F --> D
C -. "errors" .-> H[Self-healing\nauto-fix or escalate]
H .-> C
```
---
Claude Kanban orchestrates autonomous [Claude Code](https://docs.anthropic.com/en/docs/claude-code) sessions as a zero-touch build-review-ship pipeline. You describe what you want; AI brainstorms, builds, reviews, fixes, and commits. You only step in when it can't fix itself.
5 dependencies. Zero build step. One command to start.
## ◆ Quick Start
> **Prerequisite:** [Claude Code](https://docs.anthropic.com/en/docs/claude-code) installed and authenticated.
```bash
# macOS / Linux
scripts/start.sh
# Windows
scripts\start.bat
# Or manual
pnpm install && pnpm start
```
Open `http://localhost:51777`. Stop with `scripts/stop.sh` or `scripts\stop.bat`.
## ◆ Two Modes
### ◇ Single-Project (Autonomous)
Drop an `idea.md` in a folder. Walk away.
```
KANBAN_MODE=single-project
SINGLE_PROJECT_PATH=/path/to/project
AUTO_PROMOTE_BRAINSTORM=true
```
Discovery reads `idea.md` as the north star. Every 30 minutes it analyzes the codebase and creates improvement cards. Each card flows through brainstorm, spec, build, review, auto-fix, commit, push. No human needed.
```
/your/project/
idea.md ← your vision
package.json ← created by orchestrator
src/ ← built autonomously
```
**Demo included:** `sudo bash autoconfig.sh` → option 1. Watch the orchestrator build an entire project from a single idea file.
### ◇ Global (Multi-Project)
Traditional kanban. You create cards, assign project folders, control every stage.
```
KANBAN_MODE=global
PROJECTS_ROOT=~/Projects
```
```
~/Projects/
my-app/
another-project/
```
Every stage requires manual approval: spec review, build start, code review, commit.
**Human authority in both modes:** Log in anytime. Revert files, reject cards (with reasons the AI learns from), stop builds, pause the pipeline. You always override the orchestrator.
## ◆ What It Does
#### ◇ Autonomous Pipeline
Create a card, pick a project folder, and the full cycle runs: brainstorm spec, snapshot files, build, AI review (1-10), auto-fix if needed, commit on approve, rollback on reject. Card dependencies respected. Concurrency configurable.
#### ◇ Self-Healing Engine
Error scanner runs every 30s, groups failures by card, auto-fixes (2 attempts), escalates to a new card if it can't. Intelligence engine learns patterns, auto-labels cards, tunes timeouts.
#### ◇ Rejection Learning
When you reject a card, the orchestrator asks why. Your feedback is stored and injected into future brainstorm and discovery prompts. The AI adapts to your preferences over time.
#### ◇ Real-Time Board
SSE live updates. Drag-and-drop. 9-step progress visualization. Labels, search, diff viewer, inline file editing. Dark mode. WCAG 2.2 AAA. Keyboard shortcuts (`N` new, `/` search, `D` dark mode).
#### ◇ Two-Server Security
Public board on `0.0.0.0`, admin panel on `127.0.0.1` (kernel-level TCP reject). Argon2id passwords, AES-256-GCM encryption, JWT auth, CSP nonce, CSRF, token bucket rate limiting. Roles: superadmin > admin > user.
#### ◇ Production Ops
Tiered backups (5min / hourly / daily). Factory reset. Health probes (`/health`, `/health/ready`). Structured JSON logging (pino). Heartbeat watchdog. Graceful shutdown with pipeline drain.
## ◆ Deploy
| Method | Command | Use case |
|:------:|---------|----------|
| **Local** | `scripts/start.sh` | Dev, testing |
| **GCP Ubuntu** | `sudo bash autoconfig.sh` | Production (systemd, Nginx, UFW, auto-update) |
| **Docker** | `docker build -f deploy/Dockerfile -t claude-kanban .` | Containers |
| **Compose** | `cd deploy && docker compose up -d` | Self-hosting |
| **Kubernetes** | `kubectl apply -k deploy/k8s/` | Orchestrated |
See [Deployment](docs/deployment.md) for full details.
## ◆ Trust
**305 automated tests** across 5 suites. [View the report →](public/product/trust/)
| Reliability | Dependency Audit | Performance | Data Durability | Code Quality |
|:-----------:|:----------------:|:-----------:|:---------------:|:------------:|
| **168** | **27** | **28** | **35** | **47** |
```bash
pnpm test # all suites
pnpm test:reliability # single suite
```
## ◆ Architecture at a Glance
```
┌─────────────────────────────────────────────────────────┐
│ Cloudflare Edge │
│ DNS · CDN · WAF · DDoS Shield │
└──────────────────────┬──────────────────────────────────┘
│
┌─────────────┴─────────────┐
│ │
┌────▼────┐ ┌──────▼──────┐
│ Public │ │ Admin │
│ :51777 │ │ :random │
│ 0.0.0.0 │ │ 127.0.0.1 │
│ │ │ (TCP reject │
│ Board │ │ external) │
│ API │ │ │
│ SSE │ │ Control │
│ Auth │ │ Backups │
└────┬────┘ │ Users │
│ └──────┬──────┘
│ │
┌────▼───────────────────────────▼────┐
│ SQLite (WAL mode) │
│ cards · sessions · audit_log │
│ users · config · error_log │
│ learnings · checkpoints │
│ claude_usage │
└──────────────────┬─────────────────┘
│
┌──────────────────▼─────────────────┐
│ Service Layer │
│ Pipeline · Brainstorm · Review │
│ Auto-discover · Intelligence │
│ Snapshot · Git · Claude Runner │
└────────────────────────────────────┘
```
## ◆ Documentation
| | Document | What's Inside |
|-|----------|---------------|
| ◇ | [Architecture](docs/architecture.md) | Two-server design, service layer, database schema |
| ◇ | [Pipeline](docs/pipeline.md) | Build-review-ship cycle, auto-fix loop, self-healing |
| ◇ | [API Reference](docs/api.md) | 100 endpoints with request/response examples |
| ◇ | [Configuration](docs/configuration.md) | 70+ environment variables, custom prompts |
| ◇ | [Deployment](docs/deployment.md) | Local, Docker, K8s, GCP autoconfig |
| ◇ | [Security Audit](docs/security-audit.md) | Three audits, 35 findings + 72 CodeQL alerts fixed |
## ◆ Tech Stack
| Layer | Choice | Why |
|-------|--------|-----|
| Runtime | Node.js ≥ 18 | Zero-config, cross-platform |
| Server | Express | 5 deps, no framework overhead |
| Database | SQLite (WAL) | Single file, zero ops, instant backups |
| Auth | Argon2id + JWT + AES-256-GCM | Password hashing, tokens, field encryption |
| Frontend | Vanilla JS | No build step, 150KB total, ships as-is |
| Logging | Pino | Structured JSON, correlation IDs |
| CI | GitHub Actions | Matrix: Ubuntu/Windows, Node 18/22 |
## ◆ License
[Apache 2.0](LICENSE) -- [Divya Mohan](https://dmj.one)
See [NOTICE](NOTICE) for attribution requirements.
---
◆ Built with care. Ships with confidence. ◆