ClawBus
ClawBus is a minimal coordination protocol for multi-agent Claude Code — a trust layer that makes delegation, approval, and audit first-class protocol events instead of UI callbacks.
Core trick
Routes every file mutation through a typed approval-request message stored in an append-only, causally linked log.
Description
ClawBus is a minimal coordination protocol for multi-agent Claude
Code — a trust layer that makes delegation, approval, and audit
first-class protocol events instead of UI callbacks. It lowers the
trust barrier to multi-agent Claude Code by making every task,
approval, and result inspectable after the fact.
The itch came from running multiple Claude Code sessions across a
5-node cluster for months: every coordination layer I built reached
for workflow engines bigger than needed, while hand-offs still
happened by copy-pasting context between tabs and approving diffs
on vibes.
ClawBus replies with five message kinds (task / result / approval-
request / approval-decision / log), an append-only store with causal
parent links, and an adapter interface four methods long. Three
adapters ship — File, SQLite, Discord; the last lets a human ✅
reaction become an approval-decision without the worker knowing
which transport it's on. Same protocol, different transport,
identical audit trail.
Built on the Claude Agent SDK with Opus 4.7, every file mutation
routes through an approval-request — a fully inspectable message
carrying the proposed diff, severity, and rationale. The gate
doesn't pause on vibes.
MIT TypeScript, 27 tests on CI, three captured end-to-end runs
auditable in five minutes.
README excerpt
# ClawBus > **A minimal protocol turning multiple Claude Code sessions into an observable, auditable agent team.** **Claude Code is excellent per session. ClawBus makes cross-session coordination auditable without inventing a new runtime.** Claude Code is powerful as a single coding agent, but real work usually needs a team — planning, investigation, implementation, review, and human approval. ClawBus defines a minimal protocol and reference implementation so multiple Claude Code / Claude Agent SDK sessions can delegate subtasks, share context, and surface decisions that need human sign-off. This repository is the submission for **Built with Opus 4.7: a Claude Code Hackathon** (Cerebral Valley × Anthropic, 2026-04-21 – 2026-04-26). --- ## TL;DR for hackathon reviewers ```bash git clone https://github.com/tokimwc/clawbus.git cd clawbus npm install npm run build export ANTHROPIC_API_KEY=sk-ant-... npx clawbus demo ``` You should see: 1. A **Planner** agent decompose a toy task (`examples/broken-node-project` has a bug). 2. A **Worker** agent investigate and propose a patch. 3. A **Human Approval Gate** pause at the terminal asking you to approve/reject. 4. On approval, the patch is applied and `npx clawbus logs` shows the full causal timeline of messages. Everything runs locally on SQLite. No network calls except to the Anthropic API. See [`docs/judging-guide.md`](docs/judging-guide.md) for a 5-minute review checklist with branches for "no API key" vs "with API key" r