gpu-cli

Original source / Raw SKILL.md

---
name: gpu-cli
description: Safely run local `gpu` commands via a guarded wrapper (`runner.sh`) with preflight checks and budget/time caps.
argument-hint: runner.sh gpu [subcommand] [flags]
allowed-tools: Bash(runner.sh*), Read
---

# GPU CLI Skill (Stable)

Use this skill to run the local `gpu` binary from your agent. It only allows invoking the bundled `runner.sh` (which internally calls `gpu`) and read-only file access.

What it does
- Runs `gpu` commands you specify (e.g., `runner.sh gpu status --json`, `runner.sh gpu run python train.py`).
- Recommends a preflight: `gpu doctor --json` then `gpu status --json`.
- Streams results back to chat; use `--json` for structured outputs.

Safety & scope
- Allowed tools: `Bash(runner.sh*)`, `Read`. No network access requested by the skill; `gpu` handles its own networking.
- Avoid chaining or redirection; provide a single `runner.sh gpu …` command.
- You pay your provider directly; this may start paid pods.

Quick prompts
- "Run `runner.sh gpu status --json` and summarize pod state".
- "Run `runner.sh gpu doctor --json` and summarize failures".
- "Run `runner.sh gpu inventory --json --available` and recommend a GPU under $0.50/hr".
- "Run `runner.sh gpu run echo hello` then post the output".

Notes
- For image/video/LLM work, ask the agent to include appropriate flags (e.g., `--gpu-type "RTX 4090"`, `-p 8000:8000`, or `--rebuild`).


---

## Skill Companion Files

> Additional files collected from the skill directory layout.

### README.md

```markdown
GPU CLI ClawHub Skill (Stable)

Purpose: Safely run local `gpu` commands from OpenClaw/ClawHub agents with guardrails (dry‑run preview, command whitelist, budget/time caps, and clear remediation), without modifying GPU CLI itself.

What this skill does
- Executes only `gpu …` commands via a wrapper (`runner.sh`) with a strict whitelist and injection checks.
- Performs preflight checks (`gpu --version`, `gpu doctor --json`).
- Defaults to dry‑run previews; can enforce confirm + caps.
- Provides optional cost/time caps before execution; attempts cleanup (`gpu stop -y`) on timeout/cancel.
- Maps common exit codes to helpful guidance (auth, daemon restart, transient retry).

Non-goals
- No telemetry, no secret handling; uses the installed GPU CLI and your provider keys in OS keychain.
- Does not change GPU CLI behavior; all safeguards are in this skill only.

Quick usage (from an agent)
- Trigger: "/gpu" or phrases like "Use GPU CLI to …" (as configured on ClawHub)
- Example: "Use GPU CLI to run gpu status --json"
- Example: "Use GPU CLI to run gpu run python train.py on an RTX 4090"

Files
- `manifest.yaml` — ClawHub skill metadata, permissions, triggers, settings.
- `runner.sh` — Execution wrapper with guardrails.
- `selftest.sh` — Local checks for preflight and injection denial.
- `templates/prompts.md` — Curated prompts for common tasks.

Notes for publishers
- Mark channel as Stable, add logo/screenshots/demo, and link docs at `apps/portal/content/docs/ai-agent-skill.mdx`.
- Keep permissions minimal: Bash + Read, workspace‑scoped; network off for the skill itself.


```

### _meta.json

```json
{
  "owner": "angusbezzina",
  "slug": "gpu-cli",
  "displayName": "GPU CLI: Remote GPU Compute for ML Training and Inference",
  "latest": {
    "version": "1.1.1",
    "publishedAt": 1773339458169,
    "commit": "https://github.com/openclaw/skills/commit/9dbee3f8874a549c116b78b0f9ef16b77ccff592"
  },
  "history": [
    {
      "version": "1.1.0",
      "publishedAt": 1772036192971,
      "commit": "https://github.com/openclaw/skills/commit/ef2ed58dadb0975d83eaabdf32d0600e77d86135"
    }
  ]
}

```