Quickstart

This is the shortest path from an installed agent to a first useful project check.

1. Install

Use the Windows or Linux and macOS installer for a published release. It targets Claude Code and OpenCode by default. Or work from a checkout and use the checked-in agent files. Runtime discovery is not verified; see compatibility.

2. Open the project repository

Start your AI coding agent in the repository you plan to use for the hackathon. sherpa reads and writes project state under that repository's .sherpa/ directory.

3. Ask for help and initialize

/sherpa help
/sherpa init

If .sherpa/ is already present, ask for /sherpa status first and review the existing files before continuing.

4. Give it project context

Provide the challenge brief, judging criteria, time constraints, team skills, and repository context when known. If information is missing, sherpa is intended to mark it UNKNOWN rather than invent it.

5. Choose a workflow

/sherpa suggest focus on accessible tools for campus communities
/sherpa status
/sherpa next

Suggest explores and compares ideas; it does not select one. Once a person selects an idea, /sherpa plan can create a project plan. Implementation is a separate workflow and requires an approved plan.

Understand the result

Read the agent's report and any artifacts it identifies under .sherpa/. Follow the named evidence and approval requirements. For a map of decisions and phases, see the lifecycle.

If something does not work

Run sherpa doctor for management CLI diagnostics if installed. Confirm you're in the intended project, then see troubleshooting.