StarterGitHub

A place for the
work to grow.

A small starting point for working with agents.
Keep the context, the work, and what you learn together.

Start a project Read the idea
Project workflowSelect a part to explore
A questionWhat are we trying
to achieve?
Update decisions and evidence
Record the project context.Record the outcome, decisions, and constraints in the brief and relevant documents.
A working convention

A little structure.
A lot of room.

The project rarely arrives fully formed. Start with the question. Give the work somewhere to go.

Starter creates a shared foundation for code, analysis, research, reports, and presentations. A short map guides your agents. A brief holds the intent. Your tools and your structure grow with the work.

Why the environment matters

Know where
everything belongs.

Separate the project's memory, the work in progress, and the result. Explore what each part holds.

Intent, decisions, plans, and the evidence worth keeping. A short agent map points to the right document, instead of loading everything at once.

docs/

your-project/The scaffold
AGENTS.mdARCHITECTURE.md
docs/
  • design-docs/
  • exec-plans/
  • product-specs/
  • references/
  • WORKFLOW.md
  • QUALITY_SCORE.md

Maintained knowledge

No application stack chosen for you. Add code folders and dependencies when the work needs them.

A small loop.
A better next run.

Give the agent an outcome and a way to check it. Keep the useful learning in the project.

What do we know? What is still missing?

Read the brief, inspect the current artifact, and establish a baseline.

The question + the current state
Fix actionable failures and rerun affected checks. If repeated attempts fail, record what is missing and stop that loop.
How the working loop fits together

Start with
your question.

Point Codex or Claude at the repository.
Choose a starting point, then make it yours.

All the ways to get started
Prompt for your agent
Use https://github.com/spencerthomas/starter to scaffold a data analysis project in ../customer-retention. We need to understand why customers leave.

Then, begin with the work. Name the population, timeframe, and source data. Check missing values, duplicates, and joins before drawing a conclusion.

Project structure
for agent work.

How Starter organizes project context, agent instructions, and verification. Includes source references and implementation limits.

Read the essay

Built around the ideas in OpenAI’s harness engineering article, adapted for a smaller starting point.

Sources & background