Bridge Commander

As you work with AI, your attention gets fragmented — driving multiple planning tasks while overseeing multiple implementation tasks. Chat quickly becomes the wrong UX for piloting a fleet of agents.

This skill lets you use Claude Code / Codex as multiple chiefs of staff (lieutenants). You get a web UI where you work together, as work items get done by independent agent sessions on a kanban board.

Install

Just some dependencies and a new skill:

# dependencies
curl -fsSL https://kunchenguid.github.io/treehouse/install.sh | sh
curl -fsSL https://raw.githubusercontent.com/kunchenguid/no-mistakes/main/docs/install.sh | sh

# bridge-commander
npx skills add tonylampada/bridge-commander -g -y   # first /bridge-commander run clones the full tool

Quickstart

• Create an empty folder (e.g. myfleet)
• Start claude in that folder, inside tmux (not optional — the lieutenant lives in the tmux session)
• /bridge-commander
• Open the printed board URL (default http://localhost:4780/)
• Talk to your lieutenant from there — he'll guide you through the rest of the setup

Board views

The board region has three modes, toggled next to the filter (▦ / ☰ / 🧊):

• ▦ board — the kanban, as always.
• ☰ table — every live card as a sortable row (status, owner, labels, PRs, activity…); same cards, same filters, denser reading.
• 🧊 archived — a read-only browser over the archive (the append-only log of frozen card snapshots), newest first, paged in on demand. Clicking a row opens the regular card detail — body, timeline, frozen thread — where unarchive restores the card to the live board.

Filtering is one shared control across all three: the topbar text input plus the funnel popup (status / type / owner / label / updated — every dimension multi-select, OR within a dimension, AND across). Clicking a label or owner anywhere toggles it as a filter chip; the funnel badge counts what's active.

UI dev playground

node dev/ui-server.js (default 127.0.0.1:4790, --port/--host flags) serves the real ui/ against an in-memory fixture board from dev/fixtures/ — every endpoint faked, writes mutate and re-broadcast, nothing persists. Iterate on the UI with realistic gnarly states (dead lieutenants, giant cards, a paginated archive) without touching a live workspace.

Dependencies

• Node ≥ 18, tmux, git
• Claude Code, authenticated — the default agent harness
• GitHub CLI, authenticated — PR flows
• treehouse — worker worktrees (optional; falls back to git worktree)
• no-mistakes — only for no-mistakes-mode projects; the /no-mistakes skill appears after running no-mistakes init in the project
• OpenAI Codex CLI — only for --harness codex (optional)

Configuration

Per-workspace config lives in .bridge-commander/config.json:

Env knobs (set on the server process):

Lifecycle hooks

The workspace can react to card/worker lifecycle events with its own scripts: every executable file in .bridge-commander/hooks/<event>/ runs on that event (alphabetical, sequential, cwd = workspace root) with context in env — BC_EVENT, BC_CARD, BC_REPO, BC_WORKTREE, BC_BRANCH. Events: worker-done, worker-died, card-archived (fires before the worktree is released). Hooks are fire-and-forget — a failure or timeout never blocks the lifecycle; results land on the card timeline (hook-ran / hook-failed). Typical use: tearing down infrastructure a worker left running (dev containers, compose stacks) when its card finishes.

Network exposure

The board has no application-level auth — whoever reaches the bind address fully controls the board, including starting workers (running code):

• Default (recommended): loopback only (127.0.0.1).
• Private mesh (e.g. Tailscale): set host to that interface's address; a loopback listener is kept alongside. The mesh is your only auth boundary.
• Never bind 0.0.0.0.

How it works inside: ARCHITECTURE.md. The conceptual API (docs/api/overview.md) is the spec the implementation follows.