skill-system-tmux

Tmux execution support for long-running and persistent commands. Load this skill when you need to: (1) run commands expected to exceed tool timeout (over 60s), (2) start persistent servers or dev processes, (3) run TUI/interactive applications, (4) execute parallel isolated tasks in separate sessions, (5) run remote commands over SSH that must survive disconnection. Trigger phrases: 'run in background', 'start server', 'long-running', 'tmux session', 'keep running', 'persistent process', 'dev server', 'training script', 'git clone large repo', 'docker build'.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "skill-system-tmux" with this command: npx skills add arthur0824hao/skills/arthur0824hao-skills-skill-system-tmux

Skill System Tmux

Route shell commands through tmux when they risk timeout, require persistence, or need interactive terminal access. This skill provides decision heuristics and session management conventions — not tmux tutorials.

Decision Matrix

Use this matrix to decide: direct bash vs tmux.

ConditionRouteExamples
Estimated time > 60stmuxgit clone <large>, npm install, docker build, training scripts
Command does not self-terminatetmuxnpm run dev, python app.py, jupyter notebook
Requires TUI or interactive inputtmuxvim, htop, pudb, claude
Parallel isolated execution neededtmuxMultiple agents, concurrent test suites
Remote execution over SSHtmuxCommands that must survive connection drops
Atomic, non-interactive, fast (<30s)direct bashls, grep, git status, cat, mkdir
Single-shot with expected outputdirect bashpython script.py (fast), git diff, npm test (short)

Default: direct bash. Only route to tmux when a condition above is met.

Time Estimation Heuristics

Commands likely to exceed timeout:

  • Git operations: clone (large repos), push (large payloads), checkout (>1GB working tree)
  • Package install: npm install, pip install (many deps), cargo build
  • Build tools: docker build, make (large projects), webpack (production builds)
  • Data processing: Training scripts, preprocessing pipelines, large file transfers
  • CI/test suites: Full test runs on large projects

Session Naming Convention

Use project-context naming for human readability:

Format: {project}-{purpose}

Agent-created sessions (MANDATORY prefix):
  oc-{purpose}        # e.g., oc-audit, oc-verify, oc-render
  oc-{project}-{task}  # e.g., oc-skills-build, oc-subproject-test

User/infra sessions (no prefix required):
  skills-dev-server
  frontend-tests
  ml-training
  exp_runner

Rules:

  • Agent-created sessions MUST use oc- prefix — enables bulk identification and cleanup
  • Lowercase, hyphens for separators (not underscores)
  • Include project context for multi-project environments
  • Reuse well-known session names when they exist (e.g., exp_runner)
  • Never create sessions with numeric-only names (e.g., 531)

Core Patterns

Pattern 1: Fire and Forget (Background Task)

For long-running commands where you don't need real-time output:

# Kill any existing session with same name, then start fresh
tmux kill-session -t {session} 2>/dev/null || true
tmux new-session -d -s {session} bash -c '{command}'

Using interactive_bash tool:

tmux_command: new-session -d -s {session}
tmux_command: send-keys -t {session} "{command}" Enter

Pattern 2: Persistent Server

For dev servers or processes that must keep running:

tmux_command: new-session -d -s {session}
tmux_command: send-keys -t {session} "{command}" Enter

Check if running later:

tmux_command: has-session -t {session}

Pattern 3: Output Capture

Read what a tmux session has produced:

tmux_command: capture-pane -p -t {session}

For longer history (last 1000 lines):

tmux_command: capture-pane -p -t {session} -S -1000

Pattern 4: Readiness Detection

After sending a command, poll to detect completion:

  1. Capture the pane output
  2. Look for shell prompt at end of output (regex: /(\\$|>|#)\s*$/m)
  3. If prompt appears, command has finished
  4. If no prompt after expected duration, assume still running or hung

Pattern 5: Clean Shutdown

Always clean up sessions when done:

tmux_command: kill-session -t {session}

Kill-before-new pattern prevents zombie sessions:

tmux_command: kill-session -t {session}
# (ignore error if doesn't exist)
tmux_command: new-session -d -s {session}

Pattern 6: Bulk Resource Reclaim

Reclaim all agent-created sessions and detect stale resources. Run this when:

  • Machine feels sluggish or memory is high
  • Before/after long agent work sessions
  • User requests cleanup

Step 1: Identify agent sessions

tmux list-sessions -F '#{session_name} #{session_activity}' | grep '^oc-'

Step 2: Kill all agent sessions

tmux list-sessions -F '#{session_name}' | grep '^oc-' | xargs -I{} tmux kill-session -t {}

Step 3: Detect stale non-agent sessions

A session is likely stale if:

  • Last activity > 6 hours ago AND no running foreground process
  • Session has only an idle shell prompt (capture-pane shows just $ or ╰─)
# List sessions with last activity timestamp
tmux list-sessions -F '#{session_name} #{session_activity}'
# Compare #{session_activity} (epoch) against current time
# Capture pane of suspects to verify idle state

Step 4: Reclaim orphaned processes

# Find opencode processes whose parent tmux pane was killed
ps aux | grep opencode | grep -v grep
# Check if port is held by a zombie
lsof -i :{port} 2>/dev/null
# Force kill if needed (SIGTERM first, SIGKILL if unresponsive after 5s)
kill {pid} && sleep 5 && kill -0 {pid} 2>/dev/null && kill -9 {pid}

Important: Never kill sessions matching known infra patterns (exp_runner, manager, preprocess, watcher, unified) without explicit user confirmation.

Integration with Agent Tools

interactive_bash Tool

The primary interface. Pass tmux subcommands directly (without tmux prefix):

# Create session
tmux_command: new-session -d -s my-session

# Send command
tmux_command: send-keys -t my-session "npm run dev" Enter

# Read output
tmux_command: capture-pane -p -t my-session

# Kill session
tmux_command: kill-session -t my-session

bash Tool with Timeout

For commands near the timeout boundary, prefer increasing timeout parameter over tmux:

bash(command="npm install", timeout=300000)  # 5 min timeout

Use tmux only when:

  • Timeout cannot be reliably estimated
  • Process must persist beyond the current tool call
  • Interactive/TUI access is needed

Anti-Patterns

Don'tDo Instead
Route every command through tmuxUse direct bash for fast, atomic commands
Forget to kill sessions after useAlways clean up with kill-session
Use generic session names like s1Use descriptive names: project-purpose
Create agent sessions without oc- prefixAlways prefix: oc-audit, oc-verify, oc-build
Poll capture-pane in tight loopsUse 1-2s intervals between polls
Start a new session without killing old oneAlways kill-before-new for same session name
Send commands without clearing the line firstSend C-u before commands if line might have residual input
Leave sessions running after task completesRun bulk reclaim (Pattern 6) at session end

Existing Project Conventions

This skill system's sibling projects use these established patterns:

  • Session name exp_runner: Used for experiment execution (training, evaluation loops)
  • Watcher sessions: Background monitors that wake agents via tmux send-keys
  • machines.json config: tmux_session field defines per-machine session names
  • Kill-before-new: Standard cleanup pattern used across all experiment scripts

When working in a project with existing tmux conventions, follow them. This skill provides defaults for projects without conventions.

{
  "schema_version": "2.0",
  "id": "skill-system-tmux",
  "version": "1.0.0",
  "capabilities": ["tmux-route", "tmux-session-manage", "tmux-capture", "tmux-reclaim"],
  "effects": ["proc.exec"],
  "operations": {
    "route-command": {
      "description": "Decide whether a command should use direct bash or tmux, and execute accordingly.",
      "input": {
        "command": { "type": "string", "required": true, "description": "The shell command to execute" },
        "estimated_duration": { "type": "string", "required": false, "description": "Estimated duration: short (<30s), medium (30-120s), long (>120s)" }
      },
      "output": {
        "description": "Execution result or session name for tmux-routed commands",
        "fields": { "route": "string", "session": "string", "result": "string" }
      },
      "entrypoints": {
        "agent": "Apply decision matrix from SKILL.md, then execute via bash or interactive_bash"
      }
    },
    "manage-session": {
      "description": "Create, list, kill, or check tmux sessions.",
      "input": {
        "action": { "type": "string", "required": true, "description": "Action: create, kill, list, check, capture" },
        "session_name": { "type": "string", "required": false, "description": "Target session name" }
      },
      "output": {
        "description": "Session status or captured output",
        "fields": { "status": "string", "output": "string" }
      },
      "entrypoints": {
        "agent": "Use interactive_bash with appropriate tmux subcommand"
      }
    },
    "reclaim-resources": {
      "description": "Bulk cleanup of stale agent tmux sessions and orphaned processes. Kills all oc-* sessions, detects idle non-agent sessions, and reclaims zombie processes holding ports.",
      "input": {
        "scope": { "type": "string", "required": false, "description": "Scope: agent-only (default, kills oc-* only), all-stale (includes idle non-agent sessions with user confirmation)" }
      },
      "output": {
        "description": "Summary of killed sessions and processes",
        "fields": { "sessions_killed": "array", "processes_killed": "array", "ports_freed": "array" }
      },
      "entrypoints": {
        "agent": "Follow Pattern 6 (Bulk Resource Reclaim) in SKILL.md"
      }
    }
  },
  "stdout_contract": {
    "last_line_json": false,
    "note": "Agent-executed; uses interactive_bash tool for tmux operations."
  }
}

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Coding

skill-system-github

No summary provided by upstream source.

Repository SourceNeeds Review
General

skill-system-memory

No summary provided by upstream source.

Repository SourceNeeds Review
General

skill-system-router

No summary provided by upstream source.

Repository SourceNeeds Review
General

skill-system-evolution

No summary provided by upstream source.

Repository SourceNeeds Review