# Run Python Tasks in a Remote Sandbox

Running Python from an AI assistant can require setup, libraries, and safe output handling. This skill routes Claude, Codex, and Claude Code prompts to an inference.sh Python executor with common data, media, and web libraries.

## Install

```bash
npx skillstore add inference-skills/python-executor
```

## Metadata

- Status: approved
- Slug: inference-skills-python-executor
- Skillstore revision: r1
- Version status: missing
- Tree hash: fa0ddeb78e79119e0714b2b0f55c4b5cbd7a15884e01580bfb0c4581700b6168
- Author: inference-skills
- GitHub username: inference-skills
- License: MIT
- Repository: https://github.com/inference-skills/skills/tree/main/tools/utilities/python-executor/
- Ref: a25199bc7d6b82598536822d1738eb5d5f54025b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/inference-skills-python-executor
- Manifest: https://skillstore.pages.dev/api/skills/inference-skills-python-executor/manifest

## Capabilities

- Runs Python code through the inference.sh belt CLI.
- Uses common libraries such as pandas, NumPy, Matplotlib, requests, BeautifulSoup, Pillow, OpenCV, MoviePy, and trimesh.
- Creates returned output files when code saves files in the outputs directory.
- Supports web scraping and API request examples through Python HTTP libraries.
- Runs CPU-only data, image, video, 3D, document, and automation tasks.

## Use Cases

- Analyze Small Datasets: Process CSV-style data, calculate summary metrics, and save charts without local Python setup.
- Create Media Assets: Generate images, short videos, PDFs, or 3D files using pre-installed Python libraries.
- Prototype Automation Scripts: Run short Python scripts for API calls, file transformations, and repeatable utility tasks.

## Prompt Templates

### Run a Simple Calculation

```
Use python-executor to run a short Python calculation. Print the result and explain what ran.
```

### Summarize a Dataset

```
Use python-executor to analyze this table. Return summary statistics and save one chart in the outputs directory.
```

### Transform an Image

```
Use python-executor to resize and annotate the provided image. Save the processed image in the outputs directory.
```

### Build a Multi-Step Automation

```
Use python-executor to fetch approved API data, transform it with pandas, save a report, and summarize the output files.
```

## Limitations

- Requires the inference.sh belt CLI and a successful login.
- Runs CPU-only tasks and does not provide GPU or dedicated machine learning libraries.
- Timeouts are limited to one to 300 seconds.
- Code and data are submitted to an external executor, so sensitive inputs need review.

## Best Practices

- Review code before execution, especially when it reads files or makes network requests.
- Save required artifacts in the outputs directory so they are returned reliably.
- Use short timeouts first, then increase limits only for heavier jobs.

## Anti Patterns

- Do not submit secrets, tokens, or private files unless the external executor is approved for them.
- Do not run untrusted scraped code or commands copied from unknown sources.
- Do not use this skill for GPU workloads or long-running services.

## Security Audit

- Audited at: 2026-07-05T17:15:35.228\+00:00
- Summary: Most Markdown backtick findings are false positives from code fences, package names, or inline examples. The confirmed risks are documented belt commands that submit arbitrary Python to a remote executor, outbound HTTP examples, and related skill installation commands. No prompt injection text was found, but the safety claim is not verifiable from this skill file.

## Stats

- Views: 63
- Downloads: 8
- Favorites: 0
- Popularity score: 0