Ainb Fleet

ATC (Air Traffic Control) — the single operator-visible fleet supervisor, in exactly one mode: `lite` (no LLM: deterministic scan, known-transient retry policy, safety caps, evidence) or `full` (the persistent orchestrating brain). Full mode is a plain `ainb` session (Claude or Codex) running a generated policy, woken on a heartbeat (default 15 min) built from the LLM-free `fleet needs` read — which is itself HOOKS-PRIMARY (event-sourced): ATC learns ASK/WAIT/IDLE from the materialized `current_state` table, not by scraping panes, with a tmux fallback for non-Claude agents. `atc setup` installs the global Claude Code hooks into ~/.claude/settings.json. It auto-clears the safe/blocked sessions via the fleet verbs (confident ASK → broadcast; ERR → continue within a retry cap) and escalates the uncertain ones to the phone bridge. Supersedes the `daemon` skill for managed fleets. Use to provision / inspect / tear down an ATC instance, or to switch which controller owns the fleet.

Show fleet spend — per-session, per-model, per-day, and per-group USD cost rollups for every claude/codex session, sourced live from ainb's burndown analytics (which already prices every provider call). Use when you need spend visibility across a multi-session fleet, want to find the most expensive session/model, or are checking whether any session/group has crossed a configured budget cap. Budget breaches are also delivered as notifyd alerts. Default output: text tables. Pass --format json for LLM consumption.

Center control panel — enumerate every claude session that is blocked waiting on something: a user answer (AskUserQuestion fired), an API error retry, an idle assistant turn-end with no follow-up, or an explicit WAITING: marker. Returns rich JSON with signal kind + context per session. Use this when you've stepped away from the fleet and want one place to see everything that wants your attention and answer it.

Native phone bridge — relay messages two-way between a chat channel (Telegram, Slack, and/or Discord) and your ainb sessions. Inbound chat messages route to a target session (by `name:` prefix, else a conductor- first default) and are delivered via tmux send-keys; the session's reply is captured from its JSONL transcript and sent back to the chat. Config lives in ~/.agents-in-a-box/config/config.toml under [fleet.bridge.*]; tokens resolve from $ENV / keychain refs and are NEVER passed on argv. Use to run, install, or check the bridge — this is the channel ATC pages you through.

Long-running watcher that scans every claude session every 5s and auto-sends `continue` to any session whose recent tmux pane buffer matches a known API-error regex (rate_limited, overloaded_error, internal_server_error, request_timeout, socket_hang_up, fetch_failed, ECONNRESET). Use this when you want unattended recovery from transient API failures across the fleet.

Runtime-health view of the four fleet daemons — phone bridge, notifyd, ATC, and the fleet auto-continue daemon — in one table. Each row reports state (running / stopped / unknown), pid, uptime, last activity, error count, and a HEALTH reason that distinguishes a clean stop from a crash (stale heartbeat) or a pid-recycle. Use this when you need to answer "are my background daemons actually alive?" — not just installed, but serving. Default output: a fixed-width text table. Pass --format json for tooling.

Show fleet status — every claude session running on the host, merged across ainb + claude-peers broker + background jobs. Use when you need to enumerate sessions before composing an action, check the `summary` of each session, or pipe the list into jq for filtering. (Writes go via tmux by default; a `peer_id` is just an extra discovery signal / fallback channel, not a routing requirement.) Default output: text table. Pass --format json for LLM consumption.

Ordered multi-step prompts to fleet targets, ack-gated between steps via JSONL assistant-turn-end detection. Use for cycles like disconnect→reconnect→verify, or any flow where step N+1 requires step N to have completed first. The skill BLOCKS until each target's transcript shows the next assistant turn finishing OR per-step timeout fires (default 300s).

Fan out a single prompt to selected claude sessions across the fleet. Use when you need to apply the same instruction (e.g. `/clear`, `git pull`, `remote-control disconnect`) to many sessions at once. Routing: tmux send-keys by default (the broker is an opt-in fallback, toggled via AINB_FLEET_TRANSPORT). Refuses to run without an explicit targeting flag (--all, --filter <regex>, or --cwd <substring>) — no implicit fan-out.

Fleet orchestration overview — the `ainb fleet ...` Rust subcommand namespace for driving every claude session on the host. Routes to the sub-skills (ainb-spawn / standup / broadcast / sequence / needs / daemon / atc). Invoke this for an at-a-glance map of what fleet can do; reach for the specific sub-skill for the verb you want.

Spawn a coding-agent session correctly with `ainb run`: always into a git worktree, never bare into a plain checkout. Use whenever you are about to start a Claude/Codex/Gemini/Copilot session in a terminal. Guards the invocations that silently produce a `(broken)` workspace row, two agents sharing one working tree, a `--parent` flag that never fires, or a teardown that cancels itself and leaks the session.

Workflow-backed Jarvis control panel. Runs the deterministic `hangar` workflow with verb=needs (discover → enrich → prioritize), renders the Jarvis HUD from its render-ready cards, fires AskUserQuestion per blocked session, and routes each answer back via tmux send-keys (broker fallback only). Requires the workflow gate (CLAUDE_CODE_WORKFLOWS=1). If the gate is off, fall back to the prompt-driven `/ainb-fleet:needs` skill.