ai-build-kit

AI Build Kit

Site · Releases

AI Build Kit helps people who do not write code build and run tools with a coding agent. It suits team leads, operations staff, analysts, product managers and founders who are comfortable with technology. Developers new to agent-led work can use it too.

The six commands are:

• /setup-ai-build-kit prepares a project.
• /shape turns an idea or fault into a ready piece of work.
• /implement builds a ready piece and asks before merging by default.
• /setup-hosting sets up or checks how the tool runs live.
• /maintain checks the records and live copy, and runs upkeep when due.
• /what-now says where the project stands and what to do next.

Automatic merging is off by default. The masterplan records the choice and its limits under Runs on its own:.

The kit does not host tools or store project records. It is for internal tools with a manual fallback. Passing checks do not prove that software is safe or right for your purpose. The kit cannot replace a professional developer or provide legal, medical, financial or security advice. You use it at your own risk. The terms of the services you run a tool on, such as a host or a database plan, are between you and those services, and you are responsible for reading and following them.

Quick start

Choose one route for each project. For the Claude Code terminal app alone, run these commands from your project folder:

claude plugin marketplace add gwpicard/ai-build-kit
claude plugin install ai-build-kit@ai-build-kit --scope local

Start Claude Code there and type /ai-build-kit:setup-ai-build-kit.

For another agent, or a project using several agents, run:

npx skills add gwpicard/ai-build-kit

Choose the agents you use and install all the kit's skills. Then ask your agent: "Run the setup-ai-build-kit skill."

For Codex on a Mac, start each session from your project folder with python3 .agents/skills/setup-ai-build-kit/scripts/codex-with-github.py. Setup explains any extra tools your recipe needs. Manual installation and Agent Plugins are also available.

Read the AI Build Kit site for the workflow, the rules and checks, supported agents, design, releases and sources.

A sample session

This shortened example comes from a replayed session on an office noticeboard app, using Claude Code. The tool is already live and a change is ready to merge.

Person: /implement I opened the preview and the sign-in button says "Email me a sign-in link", which is right. Merge it and put it live.

Agent: All four tests and the project check pass. I've added a dated changelog line. The review has no unresolved concern, so I'm asking Claude Code to merge it. Claude Code will show you a confirmation box before the merge runs.

After the merge, the agent reads the live page:

Agent: The sign-in button at the live address now says "Email me a sign-in link". The live copy runs this change. This read only checks the button text.

Person: Priya still sees the old "Continue" button. Can you put it out again?

Agent: The live page already shows the new button. Try a full refresh in Priya's browser. I have not deployed it again.

Contributing and licence

Read CONTRIBUTING.md to report a problem or suggest a change. The kit is provided as is, without warranty, under the MIT licence.