blossomstack

Horsie — Development skill for Claude Code

Development community

An open-source managed agent harness: bring your own key and runtime.

How to install Horsie

This entry records only its repository, not the path inside it, so there is no exact command to give. Open blossomstack/horsie and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Horsie does

An open-source managed agent harness: bring your own key and runtime.

Alternatives in Development

Knowledge Work Plugins — Open source repository of plugins primarily intended for knowledge workers to use in Claude Cowork 9734 125 18 10.5k ★
Llamacoder — Open source Claude Artifacts – built with Llama 3.1 405B 7.1k ★
Microsandbox — Open-source self-hosted MicroVM sandboxes with sub-200ms startup, hardware-level isolation via libkrun 5k ★

README

horsie

**An open-source managed agent harness.**

The project is under active development.

📖 **[docs.horsie.dev](https://docs.horsie.dev)** — guides, operations, CLI reference, and how it works.

[![One session in horsie's three views, fused into a single frame: the timeline of every agent and turn, the agent graph of subagents and sub sessions, and the transcript](docs/public/img/horsie-session-views.png)](https://youtu.be/saoVBeuFrT4)

▶ **[Watch the two-minute walkthrough](https://youtu.be/saoVBeuFrT4)** — a session end to end, a sandbox on your own machine, and an agent building a workflow that another agent then runs.

Quick start

docker compose -f docker/docker-compose.yml up -d

Server and web UI on port 3789, no external database and no config file; data persists in a `horsie-data` Docker volume. Open , sign in as `admin` with the password the first boot printed, and add a provider and a model under **Settings → Models** — a fresh server has none, and sessions cannot run a turn without one.

Then give it a sandbox. On the machine holding the code you want the agent to work on:

curl -fsSL https://get.horsie.dev | sh

horsie auth login --server http://localhost:3789
horsie connect --server http://localhost:3789 --workspace .

That registers the current directory and holds a connection open, dialling *out*, so there is no port to open. It spawns one sandbox per session. Keep it running for as long as you want that machine reachable — or configure a cloud vendor in Settings and skip it entirely.

Back in the UI: **New** → pick a model and a sandbox → send a message.

The full walkthrough is [the quickstart](https://docs.horsie.dev/start-here/quickstart/).