FAQ

Is Sherpa a standalone AI model or hosted service?

No. It is a Markdown-first skill that an AI coding agent loads, with a small Python core and a separate distribution CLI.

Where does it save project information?

In .sherpa/ inside the hackathon project's repository.

Does it automatically pick an idea or approve architecture?

No. It may recommend an idea or architecture, but human confirmation is required.

Does the submission workflow submit my project?

No. It prepares and verifies materials without submitting them.

Which agents are supported?

Sherpa ships a shared Agent Skill and host manifests for Claude Code, Codex, Gemini CLI, Qwen Code, OpenCode, and Kilo Code. Runtime loading remains NOT VERIFIED for the updated adapters. The global installer currently targets Claude Code and OpenCode; other hosts use their shared skill directory or plugin/extension flow described in the agent guide.

Is /sherpa audit the same as sherpa audit?

No. The slash command routes to the final-audit agent workflow. The standalone CLI command audits recognized managed skill installation folders.

Can I install into a project directory with the installer?

The installer project-dir mode is explicitly unsupported. Use the checked-in integration files when working from a source checkout.

Can I pin any historical version?

Not reliably. A version option exists, but the current manifest format does not resolve arbitrary historical assets.

Is there a license?

The repository README says no license file is present; do not infer licensing terms.