القائمة
Manage a multi-agent tmux workspace. Status, allocation, prompt drafting, cleanup.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Pause and reset. Gather what's scattered, commit what's ready, prepare for what's next.
Bootstrap a multi-agent project. Define the meta-process, resolve every open question, close the circle.
One agent flies high. Holds strategic context, names the unnamed, never descends into implementation.
Every session has a beginning, middle, and end. Design for session boundaries instead of being surprised by them.
Propose tmux lane allocations for pending work. Status view + assignment plan + sequencing logic.
Show the current state of all tmux lanes. Fast, visual, no ceremony.
| name | tmux-lanes |
| description | Manage a multi-agent tmux workspace. Status, allocation, prompt drafting, cleanup. |
Manage a multi-agent workspace as a set of tmux panes, each one a lane with a known state, a known assignment, and a clean lifecycle.
When you're running multiple AI agents in parallel, each in its own tmux pane, the workspace itself becomes an instrument that needs tuning. Which pane is active? Which is stale? Which is ready for new work? What's each one doing?
This skill turns the tmux window into a managed workspace: visible state, deliberate allocation, clean handoffs.
Executable sub-skills: lanes-status for a quick status view, lanes-plan for allocation proposals.
Probe the workspace and present the full picture. Three commands give you everything:
# Inventory: index, size, process, PID
tmux list-panes -t <window> -F '#{pane_index} #{pane_width}x#{pane_height} #{pane_current_command} #{pane_pid}'
# Titles: what each pane is named
tmux list-panes -t <window> -F '#{pane_index} #{pane_title}'
# State: last few lines of each pane's output
for i in 0 1 2; do echo "=== Pane $i ===" && tmux capture-pane -t <window>.$i -p | tail -5; done
From the captured output, classify each pane:
Render the result as a table:
┌─────────────────┬────────┬──────────────────────┬──────────────────────────────────────────┐
│ Pane │ Size │ Title │ Status │
├─────────────────┼────────┼──────────────────────┼──────────────────────────────────────────┤
│ 0 (left) │ 162x82 │ "meta" │ Us — meta, active │
├─────────────────┼────────┼──────────────────────┼──────────────────────────────────────────┤
│ 1 (upper-right) │ 153x40 │ "Feature work" │ Stale — last session's agent, idle │
├─────────────────┼────────┼──────────────────────┼──────────────────────────────────────────┤
│ 2 (lower-right) │ 153x41 │ "Research" │ Shell — ready for new agent │
└─────────────────┴────────┴──────────────────────┴──────────────────────────────────────────┘
The goal: one glance tells you the state of the whole workspace.
Read the work queue (task list, active items, session plan), count available lanes, and propose assignments:
Pane: 0 (left)
Assignment: Meta (us)
Work: Orchestration, review, decisions
────────────────────────────────────────
Pane: 1 (upper-right)
Assignment: Coding Agent A
Work: Implement auth middleware — tests + integration (item 1, the complex one)
────────────────────────────────────────
Pane: 2 (lower-right)
Assignment: Coding Agent B
Work: API endpoint handlers for items 2-4 (same pattern repeated)
For complex layouts, a spatial view helps the operator see the whole board:
┌─────────────────────────────┬─────────────────────────────┐
│ │ │
│ META (me) │ CODING AGENT A │
│ Upper Left │ Upper Right │
│ │ │
│ Coordination, reviews, │ Auth middleware + tests │
│ architecture decisions │ (item 1, the complex one) │
│ │ │
│ Always present. │ Long-lived. Spin up now, │
│ Light context. │ independent, no blockers. │
│ │ │
├─────────────────────────────┼─────────────────────────────┤
│ │ │
│ CODING AGENT B │ FLEX │
│ Lower Left │ Lower Right │
│ │ │
│ API handlers for items │ Available for: │
│ 2-4 (same pattern) │ - Code reviews as PRs │
│ │ come in │
│ Depends on A's middleware │ - Ad hoc investigation │
│ landing first. │ │
│ │ Spin up when needed, │
│ │ idle otherwise. │
│ │ │
└─────────────────────────────┴─────────────────────────────┘
Sequencing logic:
- Upper Right (A): Spin up now. Independent, no blockers.
- Lower Left (B): Spin up after A's middleware lands. Same agent
can continue or fresh session.
- Lower Right: Hold in reserve for reviews as A and B produce PRs.
Critical path: A middleware → B handlers → integration tests.
Everything else is parallel enrichment.
Consider:
When a lane gets an assignment, draft a deployment prompt:
Recycle stale panes when the workspace needs clearing: