agentsclimarketplace

Swarmux

Skill ghillb/swarmux/.agents/skills/swarmux

Use this skill when you need tmux-first local task orchestration for coding agents - task submission, start, inspect, attach, wait, watch, reconcile, stop, and prune. Prefer for multi-task local runs requiring machine-readable control and operator tmux visibility.From its SKILL.md

Install
npx -y skills add ghillb/swarmux --skill swarmux

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 6 commands, including `Run `swarmux doctor`.` and 5 more.

SKILL.md

2.6 KB, 578 tokens by cl100k_base, as published. Nobody here has run it

Swarmux

Use swarmux as the control plane for local coding tasks.

Setup reference: docs/getting-started.md

Primary interface

Task management is the main interface.

  • submit creates a task record
  • dispatch submits and starts in one step
  • start launches a queued task
  • list shows tasks
  • show opens one task
  • attach enters a task session
  • wait blocks until a task reaches target states
  • watch streams task status and logs
  • logs prints stored logs
  • set-ref links tasks to PRs or issues
  • reconcile repairs crashed or lost sessions
  • stop, done, and fail change task lifecycle state
  • prune cleans up finished work

When to use

  • User asks to run multiple coding tasks locally with tmux visibility.
  • User wants an agent control plane with JSON outputs for task lifecycle commands.
  • User needs lifecycle control for running tasks, task recovery, or cleanup.
  • User needs backend-backed task state (files default, optional beads via bd).
  • User wants inspectable local execution instead of hidden/background orchestration.

Agent command model

  • Runtime command is provided by task payload command and executed as-is in tmux.
  • Do not assume Codex-only runtime; Codex is a common example, not a hard requirement.

Invariants

  • Prefer --output json for all machine consumption.
  • Prefer raw payload input with --json or --json-file for mutating commands.
  • Use --dry-run before real mutations when validating a payload or workflow.
  • Treat schema as the source of truth for command shape.
  • Never bypass swarmux with ad hoc tmux or git worktree commands unless you are repairing a broken session.

Workflow

  1. Run swarmux doctor.
  2. Run swarmux init.
  3. Inspect command shapes with swarmux --output json schema.
  4. Submit or dispatch tasks with raw JSON payloads or tmux-friendly flags.
  5. Use list, show, attach, wait, and watch for supervision.
  6. Use stop, done, fail, set-ref, reconcile, and prune for explicit control.

Safety

  • Inputs are validated defensively because agents hallucinate paths and identifiers.
  • logs is sanitized by default; use --raw only when needed.
  • prune is dry-run by default. Add --apply only when intentional.
  • The beads backend is optional. Default backend is files.

What ships with it: 1 file

2.3 KB alongside SKILL.md

reference/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.