tcclaviger

Localswarm — AI skill for Claude Code

AI community

Sandboxed, ticket-driven agent swarms for Claude Code on your own model endpoint.

How to install Localswarm

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

What Localswarm does

Sandboxed, ticket-driven agent swarms for Claude Code on your own model endpoint.

Alternatives in AI

CliRelay — Self-hosted AI gateway for coding CLIs — one OpenAI/Claude/Gemini/Codex-compatible endpoint, with a multi-tena 1k ★
Kibitz — Real-time decoded feed of AI agent actions — monitor multiple Claude Code & Codex sessions, see exactly what e 521 ★
Openase — Ticket-Driven Automated Software Engineering 265 ★

README

localswarm — sandboxed agent swarms for Claude Code on your own models

A Claude Code skill plus a few scripts that let one Claude Code session (the **orchestrator**, e.g. Opus or Fable) drive many Claude Code agents running on **your** model endpoint — a local server (vLLM, SGLang) or a remote gateway (your own API, Open WebUI, ...), with or without a bearer token — on any task. Work runs as a **ticket board**: a planner agent designs a workflow that fits the task, each ticket carries its own gates (commands, an independent review, human sign-off), a dispatcher keeps the slots full and opens fix tickets when a gate fails, and a final delivery ticket cleans up and hands over. Every agent runs in a `bwrap` sandbox locked to one folder.

In the project folder you will find:

tickets/ — the durable ticket registry (BOARD.md is the readable view). It stays after delivery.
workingtemp/ — every temporary process file (prompts, logs, reviews, scratch). Archived or deleted at the end.
DELIVERY.md — what was delivered, how to use it, final check results, known gaps.

Status: a working starting point extracted from a real run. Expect to tailor it (see SETUP.md).

What you get

Path	What it is
`SKILL.md`	The `/localswarm` skill: intake questions, fence, effort, pipeline, hints, monitoring
`SETUP.md`	A prompt for an LLM agent to install and tailor all of this on a new machine
`profile.env.example`	Every machine-specific setting (endpoint, format, token file, effort names, ...)
`scripts/swarm_plan.sh`	Planner agent → validated `tickets/plan.json` → independent plan review
`scripts/swarm_board.py`	Ticket registry CLI: init, import, add, list, show, set, hint, approve/reject, unblock, pack
`scripts/swarm_dispatch.py`	Works the board: starts ready tickets, runs gates, opens review/fix tickets, escalates
`scripts/swarm_run.sh`	Run one sandboxed agent (roles `impl` and `review`) or one gate command (`--exec`)