Tesslane — AI Coding Swarm Dashboard
Your coding swarm. One local workspace.
Tesslane is an open source, local-first AI coding swarm dashboard for Codex and Claude Code. Plan features, coordinate agents in Git worktrees, and review changes from one workspace.
Check License: MIT Node.js 24+ Agent Skills Skill installs Latest release
Get started · How it works · FAQ · Documentation · Contribute
Tesslane tour: a feature plan, the agent hierarchy, and a scoped agent inspector
From intention to reviewed change
Working with several coding agents creates a second job: tracking plans, questions, branches, checks and handoffs. Tesslane brings that work into one project workspace, with a visible path from a feature idea to a reviewed Git result.
Use Tesslane to coordinate a coding swarm: break a feature into dependent tickets, assign scoped work to Codex and Claude Code agents, approve execution, and inspect the resulting checks, reviews and Git commits. The Swarm Dashboard keeps planning and human review connected to the code being changed.
| You need to… | Tesslane gives you… |
|---|---|
| Turn an idea into scoped work | Feature clarification, acceptance criteria and editable, revisioned delivery tickets. |
| Know what needs your attention | A To do inbox for decisions, approvals, blockers and results. |
| Understand your swarm | Agent hierarchy, ticket dependencies, recorded interactions and scoped transcripts. |
| Keep execution under control | Per-role model routing, autonomy settings, explicit approvals and authorized checks. |
| Review the outcome | Isolated Git worktrees, independent reviews, integration evidence and a delivery report tied to the actual commit. |
| Recover interrupted work | Durable state, retained worktrees, visible conflicts and reconciliation against observed Git state. |
Local by design. The dashboard and project state run on your machine. Model requests use your native coding engines and their existing accounts. Tesslane does not provide model access or require a Tesslane account. Privacy and data flow →
Early open source release · v0.1.0. The current application is the maintained product. Native Codex delivery has prior local evidence; genuine Claude and mixed-provider inference remain unverified. CI tests deterministic behavior without model calls. Validation boundaries →
Get started
You need Git, a signed-in Codex or Claude Code CLI, and Bash or PowerShell. The Vercel skills CLI currently requires Node.js 22.20+ and npm. First launch prepares the application in a local cache, reusing Node.js 24+ or downloading a private checksum-verified runtime.
1. Install the skill
For Codex, install it across your projects:
npx skills add Lorenzo-Coslado/tesslane --skill tesslane --agent codex --global
For Claude Code:
npx skills add Lorenzo-Coslado/tesslane --skill tesslane --agent claude-code --global
These commands work in a terminal on macOS, Linux and Windows. Review the CLI's installation summary when prompted. The skills.sh badge reports installations recorded by this CLI; it is not a count of unique users, repository clones or every installation method. See installation metrics and privacy.
Install from a retained source checkout
You can also keep a complete clone in a stable location:
git clone https://github.com/Lorenzo-Coslado/tesslane.git
cd tesslane
bash scripts/install.sh codex
Use bash scripts/install.sh claude for Claude Code. On Windows, run powershell -File scripts/install.ps1 -Agent codex or use -Agent claude. This route can prepare a private Node.js runtime even when Node is not installed on your system. It does not report an installation to skills.sh.
2. Open your project and invoke Tesslane
In Codex, invoke $tesslane. In Claude Code, invoke /tesslane.
The skill opens the dashboard for your active project. If it does not appear, restart the agent. The Vercel installation includes a self-contained launcher that prepares a pinned Tesslane release on first use and reuses its local cache. A source-checkout installation uses the retained clone recorded by its installer.
You can also launch directly:
bash scripts/start.sh "/absolute/path/to/your/project"
3. Plan, review, then run
Add a feature, answer its questions and review the plan. Select your engines, autonomy and verification commands in Settings, then start execution. Inspect agents and evidence as work progresses; use the delivery report to review the resulting Git commit and apply the explicit handoff.
Writing agents require an initial Git commit and a clean source worktree. Setup commands and local-file copies follow the dashboard's approval policy. The original source branch is preserved; worktrees are not an OS security sandbox.
Full installation, updates and troubleshooting →
How it works
flowchart LR
Idea["Describe a feature"] --> Plan["Clarify and review the plan"]
Plan --> Run["Run eligible tickets in worktrees"]
Run --> Review["Checks and independent review"]
Review --> Result["Integration commit and delivery report"]
Review -->|"Changes needed"| Run
The skill opens the right project. The local engine manages policy, scheduling and durable state. Native provider adapters run Codex or Claude Code missions. The dashboard makes decisions, execution and evidence visible. Closing the browser leaves the project engine running; reinvoking the skill reopens it.
Still screenshots of the current workspace
Feature planning
Feature workspace with acceptance criteria, delivery tickets and progress
Agent hierarchy and scoped inspection
Agent hierarchy with coordinator, implementers and independent reviewer
Selected agent with its state, follow-up destination and recorded transcript
All captures use the application's labeled sample workspace. They illustrate the interface, not a live provider run.
Built for agent skills
The entry point follows the open Agent Skills specification: a focused SKILL.md, explicit environment requirements and a small instruction body. Codex presentation metadata and the selected Tesslane icon live beside it.
skills/tesslane/
├── SKILL.md # Portable instructions and metadata
├── scripts/open.mjs # Standalone installation launcher
├── agents/openai.yaml # Codex display metadata
└── assets/tesslane.png # Skill icon
The root plugin.json describes the full application as a portable plugin package. The skill can be installed through the Vercel skills CLI or from a retained full clone. The skills.sh directory and install badge are separate from OpenAI plugin-directory listing or certification. Packaging details →
Develop locally
npm ci
npm run preview -- --no-open # Sample workspace; no model requests
npm run verify # Types, format, tests, build and launch/restart
Use npm run dev -- --no-open for the regular development app. CI runs verification on macOS, Linux and Windows, with Node.js 24 and 26. Tests use real temporary Git repositories and SQLite with simulated providers; real inference probes remain opt-in.
Documentation
| Guide | What's inside |
|---|---|
| Getting started | Requirements, skill installation, updating and troubleshooting. |
| Frequently asked questions | What Tesslane does, supported engines, costs, privacy and validation. |
| Workspace guide | Features, questions, controls, verification, delivery and recovery. |
| Architecture | System diagram, source map and execution boundaries. |
| Development | Local commands, test boundaries and media reproduction. |
| Implementation status | Proven behavior and remaining validation. |
| Specification · Glossary | Product contracts and domain terminology. |
| Design system | Implemented tokens and interaction rules. |
| Privacy · Security | Local storage, provider data flow and private reporting. |
| Changelog | Public release notes. |
Exploratory work is preserved under experiments/. The archived feature-workspace prototype was not accepted for the current product and is excluded from the production build.
Contribute
Help improve Tesslane through a bug report, feature proposal, discussion or pull request. Useful starting points include reproducible provider compatibility reports, installation journeys, accessibility, documentation and regression fixes.
Read CONTRIBUTING.md and the Code of Conduct. Report vulnerabilities privately.
License
MIT © 2026 Tesslane contributors. The project's source, skill, documentation and original visual assets are open source under this license; third-party dependencies retain their own licenses. Codex and Claude Code are separate products from their respective owners. Tesslane is an independent community project.
For references, use the canonical repository, a specific release, or the metadata in CITATION.cff. A plain-text documentation index is available for readers and tools that prefer Markdown sources.