Skill-guided workflow
Reusable SPAR skills guide agents through Specify, Plan, Act, and Retain without needing a heavy framework.
Specify. Plan. Act. Retain.
spar-kit
Focused context for effective AI development.
spar-kit is a lightweight workflow that helps AI agents align with your intent, plan clearly, act with less drift, and retain durable project memory.
Install in one command
(requires Node.js)
npx @spar-kit/install
More agents
Recommended default. Works with most AI agent platforms.
Immediate next step
After install, ask your agent to use spar-init
spar-init does more than a quick setup check. Its purpose
is updating spar-kit's local tool index so your AI knows exactly which
tools, CLIs, and project commands are available in your repo.
That matters because agents work better when they do not have to guess. With a tool index, they spend less time probing the environment, use fewer tokens on tool discovery, and make more reliable decisions about how to act in your project.
The spar-init skill will quickly:
- Check that your spar-kit install is current
- Index available repo tools in
.spar-kit/.local/tools.yaml
Workflow
A simple operating loop for real repository work.
spar-kit gives every change a clear path from intent to shipped work to retained project memory.
-
Specify
Clarify goals, scope, and constraints before implementation starts.
-
Plan
Turn the change into an explicit sequence that reduces drift.
-
Act
Implement in the repo with validation tied to the plan and spec.
-
Retain
Preserve decisions and shipped behavior as durable project memory.
What you get
Just enough structure to keep agents focused.
spar-kit installs a small, repo-native workflow layer: skills to guide the work, rules to keep agents aligned, a tool index so AI knows what is available, and a specs structure for active and completed changes.
Low-context agent rules
Concise agent guidance keeps AI tools aligned to the SPAR workflow without flooding every session with instructions.
Tool index
A local tool index helps agents see and efficiently use the commands and project tools available to them.
Spec structure
Active and completed spec folders give each change a clear home from first intent to retained project memory.
Safe for existing repos: spar-kit layers in without overwriting your
existing skills, rules, or AGENTS.md.
Why spar-kit?
AI works best with focused context.
AI agents are powerful, but they are easy to derail when the context is vague, noisy, or overloaded. spar-kit gives each change a small, practical structure: clarify the intent, capture the constraints, plan the work, act against that plan, and retain what matters when the work is done.
That structure keeps the model focused on the right problem at the right time. The spec holds intent. The plan holds execution. Retain turns shipped work into project memory. Each piece has a job, so your agent does not have to rediscover the same context every session.
The result is less drift, less rework, and a cleaner path from idea to implemented change.
Further reading
Built for disciplined delivery, not feature sprawl.
spar-kit is designed for AI builders who want lightweight structure without inheriting heavyweight process. It keeps the emphasis on understanding the change, carrying a plan all the way through implementation, and retaining the context that will matter next time.
That makes it a strong fit when broader agent toolkits feel too diffuse, spec-only workflows stop short of shipped outcomes, or formal specification systems add more ceremony than day-to-day engineering work needs. The goal is not more process. The goal is better context, steadier execution, and lasting project memory.