메뉴
Deterministically (re)build the atmux cockpit + every per-team cage via `atmux cockpit rebuild`. Idempotent — safe after reboot or accidental cage drop. Per ADR-063 / ADR-135 / ADR-162.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | cockpit-rebuild |
| description | Deterministically (re)build the atmux cockpit + every per-team cage via `atmux cockpit rebuild`. Idempotent — safe after reboot or accidental cage drop. Per ADR-063 / ADR-135 / ADR-162. |
Recreates the canonical cockpit topology in one shot via the atmux cockpit rebuild verb (ported from operator dotfiles into atmux proper per ADR-063):
| Layer | What |
|---|---|
Cockpit session atmux_cockpit (ADR-135 §D1; renamed from legacy atmux_teams) | window 1 _superdriver; optional _medic / _sentinel slots when configured (both retired as auto-spawn roles per ADR-211 + ADR-212 — slots stay for back-compat); per-team viewer windows from window N+1, each self-heal-loops tmux attach into its cage's driver window |
| Per-team cages | one tmux server per team on an atmux-resolved socket (/tmp/atmux-<team>/sock, or team.json::tmuxTmpdir override; see ADR-018 + ADR-162). Each cage spawns Claude with an isolated CLAUDE_CONFIG_DIR so per-account session state and rate-limit windows don't cross-contaminate. Member panes use --permission-mode auto + bare window names per ADR-006. |
| Cage prefix | per-cage prefix is level-resolved at rebuild time per ADR-089 §C — cages nested inside a parent cage gain a different prefix than top-level team cages, so the operator can target a specific layer without ambiguity. |
Registry ~/.claude/teams/registry.json | trimmed to the canonical team set; existing emoji rosters preserved across rebuilds. |
| Crontab | scrubbed of orphan lines from stopped teams + stale-team-name blocks. Cron SHELL / PATH / TERM=xterm-256color env preamble prepended so cron's tmux invocations don't segfault on bare env. |
atmux cockpit rebuild, passing through $ARGUMENTS if non-empty.
--no-cycle — skip the cage stop+start cycle; only normalise team.json files, registry, and cockpit overlay. Use when in-flight REPL state in cages must be preserved.--force-cycle — cycle even cages with running Claude REPL processes (default behavior protects live cages).▸ normalising team.json files, ▸ trimming registry, ▸ stopping cages, ▸ starting cages, ▸ applying level-resolved cage prefix, ▸ reconciling cockpit).state/session.txt from a corrupted prior run can desync; a doctor red predates the rebuild — --no-doctor is already baked in, but the doctor row still shows on next atmux doctor).tmux attach -t atmux_cockpit if not already inside it, and prefix n/p to flip between viewer windows.bareWindowNames (ADR-006) drops the legacy __<team>__ prefix on cage windows, so the cage shows 🦀<role> directly rather than __<team>__🦀<role>.CLAUDE_CONFIG_DIR so per-account session state + rate-limit windows stay isolated. Encoded as tuiCommands.claude in each team.json — must be the inline CLAUDE_CONFIG_DIR=… claude … form, NOT a shell wrapper function (tmux execs new-session shell-commands via /bin/sh which doesn't see user-shell-defined functions).b887009 ensures rebuild applies the level-resolved prefix per ADR-089 §C (was previously clobbering the F-key chain with a legacy fixed prefix).--no-cycle to skip the destructive part if you only need to reconcile the cockpit / registry / team.json.atmux doctor is currently red on a critical row — fix the doctor row first; the rebuild's idempotency means re-running after the fix is cheap.Cockpit rebuild touches the full topology (cockpit-tier windows + per-team cages + viewers); verdict per end-state reconciliation.
Verdict-derivation rules:
--no-cycle, or cycle hit a wedged pane that needs operator's call)._superdriver window didn't spawn, OR cockpit.json read failed, OR ≥1 expected cage failed to come up after retry.--force-cycle, re-enable a paused team, address missing cage socket.Examples:
✅ /atmux:cockpit-rebuild — 4 cages up, _superdriver + 4 viewers, all cycled
⚠ /atmux:cockpit-rebuild --no-cycle — topology reconciled, 3 cages kept their in-flight REPLs
ℹ Operator: re-run without --no-cycle when ready to refresh REPL contexts
👁 🔴 /atmux:cockpit-rebuild — cockpit.json read failed (file missing at ~/.atmux/cockpit.json)
👁 Operator: run `atmux init` per ADR-200 wizard to bootstrap cockpit.json, or restore from backup
atmux cockpit rebuild surface)atmux_teams → atmux_cockpit session rename + _-prefix cockpit-role windowsbareWindowNames convention_sentinel + _medic cockpit-role auto-spawns retired (slots in cockpit.json schema kept for one-release back-compat)Unified whip skill. Verbs — run (autonomous-work nudge loop, default), cadence (tune re-arm interval), watchdog (liveness one-shot). Use /atmux:whip <verb> or `/loop /atmux:whip` for the main loop.
Business-as-usual status check + velocity-fix escalation. Reports commit cadence, rate-limits, kanban, churn per team across the cockpit (or single team from inside a cage). Auto-escalates Dormant teams to lead. `/atmux:bau [hours] [--no-fix]`. Default window 24h.
Bruh-mode unblocker — sweeps pending decisions, blockers, flags, and worktrees and pushes everything forward in one pass. Unblock all, approve all, flip all, merge all (all worktrees we worked on).
Lightweight notification skill — atmux supervisor pings teammates via `/atmux:heads-up <message>` to nudge them about new tasks, cascade unblocks, or inbox updates. Skill just acknowledges; teammate folds the nudge into next idle turn.
Unified session-continuity skill. Verbs — cont, preclear, handoff, stop. Replaces cont, preclear, handoff, full-stop. Use /atmux:session <verb> [args].
Fleet-wide diagnose + complain sweep — runs `atmux doctor` and `atmux status --json` across every enabled team, files complaints on anomalies, and takes structural fixes (rotate lead, clear member, push branch fix). Persisted host-pressure playbook (per ADR-198) is one trigger. Per ADR-077 substrate; manually invoked (the auto-spawned cockpit role was retired per ADR-212).