Tools from our own workbench

Small Claude Code tools we built for our own work, then opened up. MIT-licensed, one job each. Read the source, file an issue, send a patch.

3 reposMITgithub.com/tellship

>/ claude code · cli

agent-spend on GitHub

See where the tokens went.

A small CLI and a /spend slash command. It reads the transcripts Claude Code already writes and prints token spend by agent and by task. The CLI is offline and read-only.

slash commandofflineread-only

TypeScript · MIT

>/ claude code · mod

agent-tree on GitHub

Watch every agent and what it handed off.

A pane docked beside the transcript that shows every agent and its delegation as a live tree. Click a row for that agent’s details.

live treedocked paneno dependencies

TypeScript · MIT

>/ claude code · skill

bureaucat-tasks on GitHub

Say the task. It lands in Bureaucat.

A skill that lets Claude capture, list, update and comment on tasks in a self-hosted Bureaucat instance, through a dependency-free Python CLI you can also run yourself.

self-hostedstdlib onlyREST API

Python · MIT

>/ want one built for you?

Describe the outcome. We'll ship it.

The team behind these repos builds custom commerce and Odoo systems to spec. You keep the source there too.

Send the spec code@tellship.com