codeaholicguy/ai-devkit

AI DevKit – a local control plane for many AI coding assistants

What it is – AI DevKit is a Node‑based tool that sits in front of a collection of AI‑powered coding agents (Claude Code, Gemini‑CLI, Codex‑CLI, Cursor, GitHub Copilot, etc.). It does not replace those models; instead it gives them a shared configuration, a console UI, a local memory store, and a set of reusable “skills” that guide the agents through a disciplined engineering workflow (requirements → design → implementation → testing → review).

Key pieces

• .ai-devkit.json – a single project‑level config that reconciles the settings each agent needs (rules, API keys, MCP endpoints, etc.).
• agent console – a terminal‑based dashboard that lists running sessions, shows details, and lets you interact with any agent.
• agent send – a command‑line way to pipe prompts, logs, or stdin to a live agent, optionally waiting for a response or targeting a named group.
• @ai-devkit/memory – a SQLite‑backed knowledge base that stores project conventions, decisions, and reusable fixes. Agents can query it on demand instead of stuffing everything into each prompt.
• Composable skills – small plug‑ins such as dev-lifecycle, verify, tdd, structured‑debug, security‑review, document‑code, and simplify-implementation. They are chained together to enforce a senior‑engineer style process.

Typical workflow

1. Run npx ai-devkit@latest init – a wizard creates the config file and scaffolds directories like .claude/, .cursor/, and docs/ai/ for each phase of work.
2. Start an agent session (ai-devkit agent console).
3. Use agent send to give the agent a task (e.g., “run the tests and report back”).
4. The dev-lifecycle skill makes the agent first ask clarifying questions, then write requirement/design/planning docs under docs/ai/.
5. When implementation begins, the verify and tdd skills ensure the agent only marks a step as done after fresh test/build output.
6. At the end, review‑type skills audit the diff against the design and run a security check.

Why it matters – As developers adopt multiple AI assistants, each tool ends up with its own rule files and separate terminal windows. AI DevKit unifies them, provides a visible control surface, and adds process‑level guards (memory lookup, verification, test‑first coding) that reduce common failure modes such as:

• agents forgetting project conventions,
• premature code generation without design,
• claiming “done” without test evidence,
• scattered configuration across many files.

Installation & usage

# One‑click bootstrap
npx ai-devkit@latest init

# List sessions
ai-devkit agent list

# Open the TUI console
ai-devkit agent console

# Send a prompt and wait for a reply
ai-devkit agent send "run the tests" --id my‑claude --wait

# Store and query project knowledge
ai-devkit memory store --title "API handlers return DTOs" \
  --content "REST handlers should return response DTOs, not domain entities." \
  --tags "api,backend"
ai-devkit memory search --query "API response convention"

Extensibility – New skills can be added from a public registry (ai-devkit skill add <registry> <skill>), pulling in contributions from Anthropic, Vercel, Supabase, Microsoft, Google, etc.

License & community – Apache‑2.0, source on GitHub, docs at https://ai-devkit.com/docs, and a typical open‑source contribution workflow.

---

Bottom line – AI DevKit is a genuine, open‑source project that provides a local‑first orchestration layer for a team of AI coding agents, turning a chaotic set of assistants into a coordinated, memory‑aware, and process‑driven development environment.