Open Source

Your AI editors,
one agent system.

Portable agents, skills, and repo context that work across Claude Code, Cursor, Windsurf, Copilot, and more. One install, every editor.

$ curl -fsSL https://raw.githubusercontent.com/asambstack/ai-cli/main/install.sh | bash

Requires git and Node.js 18+. macOS and Linux.

What you get

A complete development environment that plugs into whatever AI editor you use.

6 Specialized Agents

Engineering manager, code reviewer, debugger, refactorer, feature planner, and TDD specialist. Auto-routed from natural language.

7 On-Demand Skills

TDD workflow, security review, coding standards, design patterns, git workflow, and more. Loaded only when an agent needs them.

Slash Commands

/review /debug /refactor /feature /write-tests /manager

Repo Context CLI

Static analysis scans your project and generates context files for every editor. Zero LLM calls, zero cost.

Knowledge System

Capture learnings per repo or across workspaces. Architecture, conventions, gotchas — persisted as markdown.

Visual Dashboard

Browser UI at localhost:3141 for repo health, knowledge management, and agent config editing.

Agents

Describe your task. The right agent is selected automatically.

Agent Registry

6 active
AgentPurpose
engineering-managerBreaks compound tasks into steps, delegates to specialists, parallelizes independent work
review-agentSecurity, quality, and correctness review on diffs
debug-agentRoot cause diagnosis and minimal fix
refactor-agentDead code removal, deduplication, safe consolidation
feature-agentPhased implementation plan with dependency analysis
test-agentTDD workflow — tests first, 80%+ coverage target

CLI

Scan your repo. Generate context. Every AI editor reads it.

terminal
$ cd my-project $ ai init # scan repo, pick editors $ ai learn -c gotchas "MySQL locks on bulk insert" $ ai status railsapp · Ruby on Rails · 3 editors · 0 commits behind $ ai dashboard # open browser UI
Claude Code OpenCode Cursor Windsurf GitHub Copilot Cline Aider

How it works

One source of truth, symlinked everywhere.

1

Install

One curl command clones the repo, builds the CLI and dashboard, creates symlinks.

2

Scan

ai init reads manifests, structure, and git history. Pure static analysis.

3

Generate

Produces .ai/context.md — symlinked to each editor's context file.

4

Work

Open any editor. Agents, skills, and context are available. Describe tasks naturally.

Symlink Architecture

single source of truth
~/.ai-agents-repo/ agents/*.md ────→ ~/.ai-agents-repo/agents/ skills/*.md ────→ ~/.ai-agents-repo/skills/ claude/commands/ ────→ ~/.claude/commands/*.md claude/rules/ ────→ ~/.claude/rules/common/*.md opencode/ ────→ ~/.config/opencode/opencode.json your-project/.ai/ context.md ────→ CLAUDE.md, .cursorrules, .windsurfrules, ... learnings.md Knowledge persisted across refreshes workspace.md ────→ ../workspace/.ai/context.md