Legend

Autonomous agent pipeline — type legend in any repo.

Legend watches a Slack channel for Linear issue links, then runs a 6-stage AI agent pipeline to analyze, plan, implement, review, and ship code — fully autonomously.

How It Works

Slack message with Linear issue URL
        ↓
   ┌─────────┐
   │ Analyze  │  Read-only codebase scan
   └────┬─────┘
        ↓
   ┌─────────┐
   │  Plan    │  Write implementation plan
   └────┬─────┘
        ↓
   ┌──────────────┐
   │ Review Plan   │  Approve or revise (loops)
   └────┬──────────┘
        ↓
   ┌─────────┐
   │ Develop  │  Implement in worktree, push PR
   └────┬─────┘
        ↓
   ┌──────────────┐
   │ Code Review   │  Test + lint + review (loops)
   └────┬──────────┘
        ↓
   ┌───────────┐
   │ Close Out  │  Reply in Slack, comment on Linear
   └────────────┘

Prerequisites

Legend requires the following tools installed and authenticated on your system:

| Tool | Install | Purpose | |------|---------|---------| | Node.js ≥ 18 | nodejs.org | Runtime | | Claude Code CLI | npm install -g @anthropic-ai/claude-code | AI agent engine | | GitHub CLI (gh) | cli.github.com | PR creation & reviews | | Git | Comes with most systems | Version control |

You also need:

• A Slack workspace with a bot token (Socket Mode)
• A Linear API key
• A Gemini API key (for plan review)

Installation

npm install -g legend-cli

Or try it without installing:

npx legend-cli

Quick Start

1. Set up in your project

cd your-project
legend setup

The interactive wizard will walk you through connecting Slack, Linear, GitHub, and configuring your repos.

2. Start watching

legend

Legend will watch your configured Slack channel. When someone posts a Linear issue link, the pipeline starts automatically.

3. That's it

Legend will:

• Analyze the codebase to understand context
• Write an implementation plan
• Self-review the plan (with configurable review cycles)
• Implement the changes in a separate git worktree
• Create a pull request
• Run tests, lint, and build — fixing issues automatically
• Submit a GitHub PR review
• Reply in Slack and comment on the Linear issue

Commands

| Command | Description | |---------|-------------| | legend | Start the pipeline (watches Slack) | | legend setup | Run the interactive setup wizard | | legend status | Show current pipeline status | | legend runs | List past pipeline runs | | legend config | Display current configuration | | legend reset | Clear config and re-setup |

Flags

| Flag | Description | |------|-------------| | --verbose | Show full Claude Code turn-by-turn output | | --quiet | Suppress all logs except errors and the dashboard |

Configuration

Config is stored in .legend/config.json (gitignored) in your project root. It includes:

• Workspace — project name
• Repos — one or more repos with base branch, test/build/lint commands
• Linear — API key, team ID
• Slack — bot token, app token, channel
• GitHub — authenticated user
• Gemini — API key for plan review
• Claude — model selection
• Pipeline — max review cycles, concurrency

Multi-Repo Support

Legend supports mono-repos and multi-repo workspaces. During setup, it auto-discovers git repos in your working directory and lets you configure each one independently.

Architecture

bin/legend.ts              — CLI entry point (Commander)
src/cli/                   — Setup wizard, dashboard, sub-commands
src/config/                — Config schema and store
src/agents/runner.ts       — Claude Code subprocess invocation
src/agents/prompts/*.md    — Per-stage prompt templates
src/pipeline/              — Orchestrator state machine, worktree, queue
src/integrations/          — Slack, Linear, GitHub API clients
src/utils/                 — Logger, pre-flight checks, theme

Pipeline Artifacts

Each run persists artifacts to .legend/runs/<ISSUE-ID>/:

• analysis.md — codebase analysis
• plan.md — implementation plan
• review.md — plan review feedback
• code-review.md — code review feedback

Development

git clone https://github.com/akashmunshi/legend-cli.git
cd legend-cli
npm install
npm run dev        # Run with tsx (no compile step)
npm run build      # Compile to dist/
npm link           # Make `legend` available globally for testing

License

MIT — see LICENSE.

Built by Akash Munshi