Installation

spp ships as a Claude Code plugin.

Requirements

• Claude Code — spp runs as a plugin inside it.
• Python 3.11+ — for the runner scripts spp generates (scoring, splitting).
• A task model — any OpenAI-compatible endpoint or local server the produced prompt runs against.

Install from the marketplace

From inside Claude Code:

/plugin marketplace add JayLBean/supervised-prompt-producer
/plugin install spp@supervised-prompt-producer

This registers the spp plugin and its /spp:run entrypoint.

Local development

To run an unreleased or locally-modified version:

git clone https://github.com/JayLBean/supervised-prompt-producer.git
cd supervised-prompt-producer
claude --plugin-dir ./

--plugin-dir ./ loads the plugin from the working tree, so local edits apply on the next session.

Verify

Confirm the entrypoint is available:

/spp:run

With no task name, the consultation begins and asks for one. If the command is missing, re-check the install step, or that --plugin-dir points at the repo root.

What spp writes

spp works inside a spp/<task-name>/ directory — plan, baseline, splits, iteration artifacts, frozen prompt, and report. Nothing is written outside it without your approval, and the sacred test set is never read until gate G5.

Start using spp →