AI DevKit

English | 中文

The control plane for AI coding agents.

AI DevKit gives Claude Code, Codex CLI, Gemini CLI, opencode, Pi, Cursor, GitHub Copilot, Devin, and other coding agents one local-first operating layer: one config, one console, local memory retrieval, cross-agent communication, and composable engineering skills led by dev-lifecycle.

• One config for every agent — .ai-devkit.json reconciles setup across the coding tools your team uses
• One console for running sessions — agent console is a live TUI dashboard for supervising local agents across providers
• Cross-agent communication — agent send lets you route prompts, logs, and test output to running agents
• Memory retrieval without context bloat — @ai-devkit/memory stores decisions, conventions, and fixes in local SQLite so agents search when needed instead of carrying everything in every prompt
• Composable engineering skills — dev-lifecycle, verify, tdd, review, debugging, security, docs, and simplification skills combine into reliable workflows

The future is many AI coding agents. AI DevKit is the layer that makes them manageable.

Run npx ai-devkit@latest init and your project gets:

Who this is for

Developers whose AI coding setup has grown from one assistant into a small, messy team of agents:

• multiple terminals with no shared control surface
• separate CLAUDE.md / .cursor/rules / AGENTS.md / MCP setup per tool
• no easy way to send context, logs, or follow-up work to a running agent
• the agent forgetting yesterday's conventions
• "I've successfully implemented the feature" with a red build
• the agent diving into code without a plan and producing the wrong thing

Before AI DevKit, your agents are powerful but scattered. After AI DevKit, they have shared setup, a control surface, searchable memory, communication paths, and reusable skills that travel with your repo without bloating every prompt.

Start in 30 seconds

npx ai-devkit@latest init

One wizard. Pick your agents, install the control-plane pieces you need, and give every tool the same operating model. It writes project-local files you can review and commit. Re-run it whenever your agent list or workflow changes.

Here's what lands in your repo:

your-project/
├── .ai-devkit.json              # single source of truth (re-run init anytime)
├── .claude/                     # or .cursor/, .codex/, etc. per agent you picked
│   ├── skills/                  # dev-lifecycle, verify, memory, tdd, ...
│   └── settings.json            # MCP servers wired up (incl. @ai-devkit/memory)
└── docs/ai/
    ├── requirements/            # phase 1 — what to build, why
    ├── design/                  # phase 2 — how it'll be built
    ├── planning/                # phase 3 — task-by-task plan
    ├── implementation/          # phase 4 — execution notes
    └── testing/                 # phase 5 — coverage strategy

Operate agents like infrastructure

AI DevKit ships a agent control plane for everyday multi-agent work:

# List running sessions across providers
ai-devkit agent list

# Open the live terminal UI
ai-devkit agent console

# Send a prompt to a running session and wait for the response
ai-devkit agent send "run the tests and report back" --id <agent-name> --wait

# Pipe multi-line output into a running session
npm test 2>&1 | ai-devkit agent send --id <agent-name> --stdin

# Send a prompt to a saved group of agents
ai-devkit agent send "review this branch for release risk" --group reviewers

# Pipe a session through Telegram — operate your agent from your phone
ai-devkit channel start telegram --agent <agent-name> --daemon

Use this when work spans long-running agents, multiple providers, scheduled checks, review loops, or remote control from another channel.

Add memory without bloating context

AI DevKit memory is local SQLite knowledge for project decisions, coding conventions, and reusable fixes. Agents retrieve it when a task needs context instead of carrying every fact in every prompt.

# Store a reusable project convention
ai-devkit memory store \
  --title "API handlers return DTOs" \
  --content "REST handlers should return response DTOs instead of domain entities." \
  --tags "api,backend" \
  --scope "repo:codeaholicguy/ai-devkit"

# Search before related work
ai-devkit memory search --query "API response convention"

Compose engineering workflows with skills

The control plane is useful on its own. For larger or riskier changes, AI DevKit also installs composable skills that make agents behave more like an engineering team.

dev-lifecycle is the anchor skill. It guides the agent through requirements, design, planning, implementation, testing, and review. Other skills plug into that flow:

• memory retrieves relevant project knowledge without stuffing all context into the session
• verify blocks completion claims without fresh test or build evidence
• tdd pushes test-first implementation when behavior changes
• structured-debug keeps debugging reproducible instead of guess-and-patch
• security-review, document-code, and simplify-implementation add focused review passes when the task needs them

Get the full engineering workflow stack

Save templates/senior-engineer.yaml locally and run:

ai-devkit init --template ./senior-engineer.yaml

Bundles the built-in skills with curated additions from Anthropic, Vercel, and others: TDD, frontend design, webapp testing, doc co-authoring, React best practices, security review, and more.

A feature, end-to-end

You:    Use the dev-lifecycle skill to start requirements for OAuth login with Google

Agent:  Searches memory for prior auth conventions. Asks clarifying
        questions about scope, users, success criteria. Drafts
        docs/ai/{requirements,design,planning}/feature-oauth-login.md
        in a feature worktree. Stops before coding.

You:    Ask for a design review of feature-oauth-login

Agent:  Audits the design doc against the requirements. Flags gaps,
        proposes fixes — before any code gets written.

You:    Ask it to execute the implementation plan

Agent:  Works the planning doc task-by-task. Updates progress after
        each task. The `verify` skill blocks a task from being
        marked done without fresh test/build output.

You:    Ask for a code review

Agent:  Audits the diff against the design doc — scope creep,
        missing tests, edge cases the requirements named —
        before you push.

What changes in agent behavior

The flow above is powered by nine built-in skills, each addressing a failure mode developers see in real AI coding sessions:

Need more? ai-devkit skill add <registry> <skill> pulls from 30+ publishers — Anthropic, Vercel, Supabase, Microsoft, Google.

Works across coding agents

One .ai-devkit.json configures all of them. Add a new agent to your team without rewriting your rules.