# tao-run-deft-aoi-cosmos3

tao-run-deft-aoi-cosmos3 executes iterative training workflows for Cosmos3 visual inspection models. Designed strictly for binary OK or NG verification.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf (Devin Desktop), Claude.ai · ★ 4106 repo stars
- Author: NVIDIA Corporation
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/NVIDIA/skills/tree/HEAD/skills/tao-run-deft-aoi-cosmos3/SKILL.md
- Updated: 2026-10-23
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/tao-run-deft-aoi-cosmos3

## What it does

This workflow manages recurring training passes for NVIDIA Cosmos Reason 3 checkpoints. It uses Nano unless you specifically pick Edge or Super. The engine measures baseline accuracy against Benchmark and Proxy splits, gathers challenging real image pairs, creates synthetic faults using AnomalyGen, and formats JSON batches for Cosmos Framework LoRA SFT.

State transitions persist on disk inside `deft_state.json`. Workload orchestration relies on standard execution verbs: job submission, status polling, reading logs, and job cancellation. The agent reads durable stage records instead of guessing progress from terminal prose.

Responses must remain binary: the model produces only `OK` or `NG` labels. Running the tool requires the full TAO skill-bank tree mapped to `TAO_SKILL_BANK_PATH`, alongside host Python containing `pyarrow`, `yaml`, and TOML parsing support.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tao-run-deft-aoi-cosmos3 -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/tao-run-deft-aoi-cosmos3/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/tao-run-deft-aoi-cosmos3/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tao-run-deft-aoi-cosmos3 -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/tao-run-deft-aoi-cosmos3/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tao-run-deft-aoi-cosmos3 -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/tao-run-deft-aoi-cosmos3/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install NVIDIA/skills skills/tao-run-deft-aoi-cosmos3
```

2. **Manually: put the files in place**
- File: `.github/skills/tao-run-deft-aoi-cosmos3/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tao-run-deft-aoi-cosmos3 -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/NVIDIA/skills --path skills/tao-run-deft-aoi-cosmos3
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/tao-run-deft-aoi-cosmos3/SKILL.md` — The legacy .windsurf/skills/ path still works

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/NVIDIA/skills/tree/HEAD/skills/tao-run-deft-aoi-cosmos3/SKILL.md

## When to use it
- orchestrate Cosmos3 LoRA SFT training passes for PCB inspection
- synthesize defect pairs through AnomalyGen tooling
- mine failure cases from evaluation splits
- manage multi-step state persistence via deft_state.json

## Example prompts
- Run Cosmos3 DEFT AOI on the current board dataset using Nano
- Start CR3 AOI loop with the Edge checkpoint
- Improve Cosmos3 PCB inspection with bare OK/NG outputs

## Similar
- [nvidia-cosmos-video](https://affpapa.org/en/nevibecode/skills/nvidia-cosmos-video.md) — skill · 132 installs
- [run-github-project](https://affpapa.org/en/nevibecode/skills/run-github-project.md) — skill · 829 installs
- [run-workflow](https://affpapa.org/en/nevibecode/skills/run-workflow.md) — skill · 250 installs
- [gradle-run](https://affpapa.org/en/nevibecode/skills/gradle-run.md) — skill · 1100 installs
- [cosmos-dbt-core](https://affpapa.org/en/nevibecode/skills/cosmos-dbt-core.md) — skill · 1000 installs
- [run-release](https://affpapa.org/en/nevibecode/commands/run-release.md) — command · ★ 907 repo stars

From the same set: [nvidia/skills](https://affpapa.org/en/nevibecode/by/nvidia/skills.md) — 312 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
