メニュー
Run interactive CLIs (python REPL, terraform apply) and long-running background tasks in isolated tmux sessions. Scrape pane output and manage session lifecycle. Use when a command needs interactive input, runs for a long time, or requires a TTY.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/jingkaihe/skills --skill tmuxコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
SOC 職業分類に基づく
Use whenever working with Gmail, Google Calendar, Google Contacts, or Google Drive through the google-workspace MCP server. Covers the maco Python wrapper workflow, tool discovery, account selection, Gmail queries, attachments, event timing, Drive exports, and safe mutations.
Run AI agents and arbitrary code in ephemeral micro-VMs with VM-level isolation, network allowlisting, and host-side secret injection. Use for Matchlock CLI and SDK workflows.
Manage scheduled, background agentic tasks. Use when users ask to create, list, inspect, or delete schedules, recurring agent tasks, reminders, or background dispatch of natural-language instructions.
| name | tmux |
| description | Run interactive CLIs (python REPL, terraform apply) and long-running background tasks in isolated tmux sessions. Scrape pane output and manage session lifecycle. Use when a command needs interactive input, runs for a long time, or requires a TTY. |
Run interactive CLIs and long-running tasks in isolated tmux sessions using a dedicated socket (-L llm-agent → /tmp/tmux-llm-agent) to avoid interfering with the user's own tmux. Never omit -L llm-agent.
Create session:
tmux -L llm-agent new-session -d -s <name> '<command>'
Send input (interactive prompts, REPL input):
tmux -L llm-agent send-keys -t <name> '<text>' Enter
tmux -L llm-agent send-keys -t <name> C-c # interrupt
tmux -L llm-agent send-keys -t <name> C-d # EOF / exit REPL
Capture pane output:
tmux -L llm-agent capture-pane -t <name> -p # visible screen
tmux -L llm-agent capture-pane -t <name> -p -S -<N> # last N lines of scrollback
tmux -L llm-agent capture-pane -t <name> -p -S - # entire scrollback
Pipe through grep/tail for targeted extraction:
tmux -L llm-agent capture-pane -t <name> -p -S -100 | grep -i 'error\|fail\|warn'
tmux -L llm-agent capture-pane -t <name> -p -S -50 | tail -20
Session management:
tmux -L llm-agent list-sessions
tmux -L llm-agent has-session -t <name> 2>/dev/null && echo "running" || echo "finished"
tmux -L llm-agent kill-session -t <name>
tmux -L llm-agent kill-server # kill ALL agent sessions
By default the session closes when the command exits, so use remain-on-exit to keep the pane alive for scraping final output:
tmux -L llm-agent new-session -d -s my-build
tmux -L llm-agent set-option -t my-build remain-on-exit on
tmux -L llm-agent send-keys -t my-build 'make build 2>&1' Enter
sleep 5
tmux -L llm-agent capture-pane -t my-build -p -S -30 # check progress
# when done, grab final output and clean up
tmux -L llm-agent capture-pane -t my-build -p -S -
tmux -L llm-agent kill-session -t my-build
If you don't need final output, the simpler form works — the session auto-closes on exit:
tmux -L llm-agent new-session -d -s my-build 'make build 2>&1'
tmux -L llm-agent has-session -t my-build 2>/dev/null && echo "running" || echo "done"
tmux -L llm-agent new-session -d -s repl 'python3'
sleep 1
tmux -L llm-agent send-keys -t repl 'import os; os.getcwd()' Enter
sleep 1
tmux -L llm-agent capture-pane -t repl -p -S -10
tmux -L llm-agent send-keys -t repl 'exit()' Enter
tmux -L llm-agent new-session -d -s test-unit 'pytest tests/unit'
tmux -L llm-agent new-session -d -s test-integ 'pytest tests/integration'
tmux -L llm-agent list-sessions
tmux -L llm-agent capture-pane -t test-unit -p -S -20
Use the bundled scripts/wait-for.sh to poll a pane until a pattern appears (or timeout). It uses the -L llm-agent socket automatically and -J to join wrapped lines.
bash <skill-dir>/scripts/wait-for.sh -t <name> -p 'Apply complete' -T 60
bash <skill-dir>/scripts/wait-for.sh -t <name> -p 'Enter a value' -F # fixed string
bash <skill-dir>/scripts/wait-for.sh -t <name> -p 'ready' -i 1 -l 500 # custom interval/depth
Options: -t target, -p pattern, -F fixed string, -T timeout (default 30s), -i poll interval (default 0.5s), -l scrollback lines (default 1000). On timeout exits non-zero and dumps the last N lines to stderr.
Example — wait for terraform prompt, respond, then wait for completion:
tmux -L llm-agent new-session -d -s tf 'terraform apply'
bash <skill-dir>/scripts/wait-for.sh -t tf -p 'Enter a value' -T 120
tmux -L llm-agent send-keys -t tf 'yes' Enter
bash <skill-dir>/scripts/wait-for.sh -t tf -p 'Apply complete' -T 300
tmux -L llm-agent capture-pane -t tf -p -S -
tmux -L llm-agent kill-session -t tf
-L llm-agent on every tmux command — isolates agent sessions from the user's personal tmux.Session "<name>" started. To attach: tmux -L llm-agent attach -t <name>
To capture output: tmux -L llm-agent capture-pane -t <name> -p -S -
terraform-plan, django-server).2>&1 to capture errors in the pane scrollback.tmux -L llm-agent set-option -t <name> history-limit 50000