菜单
Deploy and manage Ollama LLM on fleet hosts. Use when user mentions ollama, LLM, AI model, or language model deployment.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Almanac reference for dates, holidays, time zones, business days, and astronomical events. Use when checking if a date is a business day, calculating date differences, converting time zones, looking up holidays, or checking for astronomical events like solstices, equinoxes, and moon phases. Also for seasonal calculations, workday schedules, and calendar arithmetic.
Toyota Production System (TPS) and Toyota Way for software engineering — lean thinking, waste elimination, root cause analysis, and continuous improvement. Use when identifying waste (muda/muri/mura), running blameless postmortems, doing 5 Whys, writing A3 problem statements, or connecting lean philosophy to TRIZ contradiction resolution.
Local multilingual translation via Ollama — 119 languages, no API key, no network, no data leaves the machine. Use when translating to/from Japanese, Chinese, Korean, French, Spanish, German, Portuguese, Russian, Arabic, or any other language. Also for localization, i18n text, and roundtrip validation. Qwen3:32b on Apple Silicon — faster and cheaper than cloud APIs.
TRIZ contradiction matrix oracle for software and systems design. Use when improving one quality worsens another — speed vs correctness, isolation vs sharing, autonomy vs control, simplicity vs completeness. Trigger phrases: "tradeoff", "tension between", "can't have both", "worsens when", "at the cost of", "faster but less reliable", "more isolated but harder to share". Not for bugs, merge conflicts, or tool choices. Prerequisite: triz binary must be installed (run `triz doctor` to verify).
Deploy and manage software on Linux/BSD fleet hosts (Pis, NAS, workstations) via Ansible. Use when user mentions deploy, ansible, fleet, hosts, inventory, or deployment.
Use when operating, debugging, or deploying the elixir-gastown Elixir/BEAM fleet — cluster federation, release builds, worker deploys, seccomp fixes, sname/name distribution issues, rate limiter, or any gastown service on Pi workers or coordinator nodes.
| name | ollama-deploy |
| description | Deploy and manage Ollama LLM on fleet hosts. Use when user mentions ollama, LLM, AI model, or language model deployment. |
| compatibility | opencode |
| metadata | {"tools_needed":"Bash, SSH, WebFetch, Glob, Grep, Read, Write","triggers":"ollama, LLM, AI model, language model"} |
Use this skill for deploying and managing Ollama (local LLM runtime) on fleet hosts.
ollama-linux-arm64.tgzollama-linux-amd64.tgzfile /usr/local/bin/ollamaOllama binary is ~1.6GB. Download once locally, then distribute:
# Download to controller
curl -fsSL https://github.com/ollama/ollama/releases/download/v0.13.1/ollama-linux-arm64.tgz -o /tmp/ollama.tgz
# Distribute via Ansible copy (parallel, fast LAN)
ansible <hosts> -i inventory -b -m copy -a "src=/tmp/ollama.tgz dest=/tmp/ollama.tgz"
# Extract on hosts
ansible <hosts> -i inventory -b -m shell -a "tar -xzf /tmp/ollama.tgz -C /usr/local"
After installing Ollama binary, pull models:
ansible <hosts> -i inventory -b -m shell -a "/usr/local/bin/ollama pull kimi-k2.5:cloud"
ansible <hosts> -i inventory -b -m systemd -a "name=ollama state=started enabled=yes"
ansible-playbook -i inventory 02-ollama.yml
ansible <hosts> -i inventory -m shell -a "/usr/local/bin/ollama list"
ansible <hosts> -i inventory -m systemd -a "name=ollama state=started"
ollama pull as _ollama userollama pull as ubuntu user or directly without become_userjournalctl -u ollama -n 50file /usr/local/bin/ollama$HOME/rust-town/irclaw/ansible/02-ollama.yml