Guide · Agent skill
How to use the Lean Build agent skill (lean-build)
Updated · written with AI from the project's own documentation and checked against our catalogue data
Lean Build is an agent skill, a set of instructions that an AI coding agent loads when a task matches it. This guide explains what the lean-build skill changes about feature work, how to install it from the ZIP on our page, how to trigger it with plain requests, and what to watch for. By the end you will know how to ask an agent for a new behavior, product slice or integration in an existing repository without getting extra modes, config or abstractions you did not request.
Lean BuildJuliusBrussee/caveman · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Get the lean-build ZIP from our page and unzip it. It contains SKILL.md and agents/openai.yaml.
- 2Choose an install location. Use ~/.claude/skills/ for all projects or .claude/skills/ inside one repository. Alternatively upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm the folder layout. Make sure the lean-build folder with SKILL.md sits directly inside the skills folder you chose.
- 4Open your existing repository. Start the agent in the codebase where the feature will live, since the skill relies on tracing and reusing current code.
- 5Describe the feature with its boundaries. Write the request with the outcome, the layers involved if known, and what is out of scope.
- 6Review the result. Check the acceptance proof and the short report of material omissions before merging.
What the lean-build skill makes the agent do
The skill targets feature work with a high risk of overbuilding. When it is installed, the agent is pushed to work in a specific order before writing code: define what success looks like and what is explicitly out of scope, trace the request through the layers that own the relevant behavior, and reuse existing code where it fits.
It discourages adding modes, providers, configuration, extensibility or polish unless the acceptance criteria need them. When a dependency, service, config change or migration is added, the agent is expected to state the material tradeoff. Once the acceptance criteria pass, the agent is told to stop and report only material omissions and the trigger.
- Derives observable acceptance criteria and non-goals from your request and the repository
- Follows the entry point through the layers that own the behavior
- Delivers an end-to-end path across the responsible layers rather than a single-file patch
- Refactors when a patch would duplicate behavior, weaken ownership or hide the root cause
- Runs a focused proof and stops when acceptance passes
Install the lean-build skill from the ZIP
Download the lean-build ZIP from our page and unzip it. The skill folder contains two files: SKILL.md and agents/openai.yaml. Then put the folder in one of the locations below.
For a personal install that applies to all your projects, place it in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder so it travels with the repository.
If you use the Claude apps instead of a folder on disk, you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/lean-build/
- Project only: .claude/skills/lean-build/ inside your repository
Trigger the skill with natural requests
You do not need a special command. The skill's description says it applies to new behavior, product slices and integrations where repository reuse, strict scope and an explicit stop condition matter, so describe your task in those terms. Mentioning the existing codebase and the boundaries you care about helps the agent match the skill to the request.
Requests that work well name the feature, the repository and the limits. Requests that suggest a greenfield or throwaway task are less likely to match, and the skill would push toward narrow scope even if you wanted freedom.
- "Add a CSV export to the reports page in this repo, reuse the existing export code, and stop once the tests for the export pass."
- "Add Stripe webhook handling for refunds to the current billing service. Keep it scoped to refund events and don't add a generic event system."
- "Implement the invite-by-email slice in this codebase. Trace it through the existing API and service layers and avoid new config."
Check what the skill needs and how it runs
The source lists no requirements. It does not run scripts and does not use the network. The skill is a set of instructions, so it adds guidance to the agent rather than executing code of its own.
The skill assumes an existing codebase, because it asks the agent to trace the request and reuse current code. It also says to preserve safety, but it does not provide security checks, so it is not a substitute for a security review.
- Requirements: none listed
- Runs scripts: no
- Uses network: no
- Not a security review: review security-sensitive changes separately
Tips for getting the best output
State your acceptance criteria in the request, since the skill derives them from the request and repository. Clear observable outcomes give the agent a stop point. If you already know what is out of scope, say so, because the skill asks the agent to list non-goals explicitly.
Expect a short report at the end. The skill asks for material omissions and the trigger only, so read those first and ask follow-up questions if something important is missing.
- Name the outcome you can observe, such as a test passing or a screen behaving a certain way
- List what is out of scope to make non-goals explicit
- Point to the existing module or layer if you know where the behavior belongs
What people use it for
- Adding a new behavior to an existing service while reusing its current seams
- Building a narrow product slice that touches API, service and storage layers in one coherent path
- Integrating a third-party service into a codebase without adding a generic plugin system
- Keeping an agent from adding unrequested config flags or provider abstractions
Frequently asked questions
How do I install the lean-build skill?
Download the lean-build ZIP from our page and unzip it. Place the lean-build folder in ~/.claude/skills/ for all projects or in .claude/skills/ inside one repository. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Does lean-build require anything else to run?
The source lists no requirements. The skill does not run scripts and does not use the network, so it works as instructions for the agent.
Is lean-build free to use?
The skill is published under the Apache-2.0 licence, as listed on this page. Check the licence file in the source repository for the full terms.
Can I use lean-build for a new project from scratch?
It is not a good fit for greenfield projects or quick throwaway scripts where broad design freedom is wanted. The skill assumes an existing codebase and pushes toward reuse and narrow scope.
Does lean-build do a security review?
No. The skill says to preserve safety but does not provide security checks, so security-sensitive changes need a separate review.
How is lean-build different from asking an agent to just build the feature?
Without it, agents may add extra abstractions, modes or config. Lean-build tells the agent to define acceptance and non-goals first, reuse existing code, and stop once acceptance passes.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Lean-build suits developers and teams adding behavior, slices or integrations to an existing repository who want narrow, reuse-focused changes. It is a poor fit for greenfield work, throwaway scripts or tasks without repository context, and it is not a security tool.
Download the skill