菜单
Provision, manage, and terminate RunPod GPU instances for LLM training. Use when user says "spin up GPU", "create RunPod instance", "terminate pod", "check GPU status", "provision training server", or needs cloud GPU resources.
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
File-based inter-agent messaging. Check inbox, send bugs/requests to other projects, acknowledge resolved issues. Use for cross-project agent communication.
Search arXiv for academic papers. Use when user says "find papers on", "search arxiv", "look up research", "academic papers about", "what papers exist on", "use your arxiv tool", or asks about scientific/ML research topics.
Step back and critically reassess project state. Use when asked to "assess", "step back", "fresh eyes", "check alignment", "sanity check", "health check", or "evaluate what's working". Offer to run after major changes (don't auto-run).
Free web and local search via Brave Search API. Use when user says "brave search", "search with brave", "brave web search", "brave local search", "local search", "find businesses near", or "near me".
Submit code review requests to multiple AI providers (GitHub Copilot, Anthropic Claude, OpenAI Codex, Google Gemini) and get patches back. Use when user says "code review", "review this code", "get a patch for", or needs AI-generated unified diffs for code fixes.
Fetch up-to-date library documentation from Context7 API. Use when user asks "how do I use this library", "show me docs for", "library documentation", "API reference for", or needs current documentation for any code library.
| name | runpod-ops |
| description | Provision, manage, and terminate RunPod GPU instances for LLM training. Use when user says "spin up GPU", "create RunPod instance", "terminate pod", "check GPU status", "provision training server", or needs cloud GPU resources. |
| allowed-tools | Bash, Read |
| triggers | ["spin up GPU","create RunPod","terminate pod","GPU instance","provision server","check pod status","RunPod management"] |
| metadata | {"short-description":"RunPod GPU instance management","project-path":"$RUNPOD_OPS_REPO (set via env; defaults to GitHub clone)"} |
Manage RunPod GPU instances for LLM training and inference.
Self-contained skill - auto-installs via uv run from git (no pre-installation needed).
# Via wrapper (auto-installs from GitHub on demand)
.agents/skills/runpod-ops/run.sh list-instances
# Create an instance
.agents/skills/runpod-ops/run.sh create-instance 70B --hours 4
# Monitor an instance
.agents/skills/runpod-ops/run.sh monitor <pod-id>
# Terminate an instance
.agents/skills/runpod-ops/run.sh terminate <pod-id>
| Command | Purpose |
|---|---|
create-instance | Create GPU pod optimized for model size |
list-instances | Show all running pods with status/cost |
monitor | Live monitoring of pod metrics |
terminate | Safely terminate a pod |
estimate-cost | Estimate training cost before creating |
optimize | Find optimal GPU config using benchmarks |
start-training | Start a training job on RunPod |
serve | Start inference server on RunPod |
.agents/skills/runpod-ops/run.sh create-instance 70B --hours 4
### Estimate Cost
```bash
.agents/skills/runpod-ops/run.sh estimate-cost 70B --hours 8
.agents/skills/runpod-ops/run.sh monitor <pod-id>
.agents/skills/runpod-ops/run.sh terminate <pod-id>
| Variable | Required | Description |
|---|---|---|
RUNPOD_API_KEY | Yes | RunPod API key |
# 1. Estimate cost
.agents/skills/runpod-ops/run.sh estimate-cost 70B --hours 8
# 2. Create pod
.agents/skills/runpod-ops/run.sh create-instance 70B --hours 8
# 3. Monitor or SSH into pod
.agents/skills/runpod-ops/run.sh monitor <pod-id>
# 4. Run training...
# 5. Terminate when done
.agents/skills/runpod-ops/run.sh terminate <pod-id>
After training completes, log the lesson:
memory-agent learn \
--problem "Training Qwen3-70B on RunPod" \
--solution "Used A100-80GB, 8 hours, cost $24.72. Config: lr=2e-5, batch=4"