ZenMux Image Gen (Nano Banana 2)

Generate images from text prompts using ZenMux API (Vertex AI protocol with Gemini models). Use when: (1) User wants to generate/create images from text desc...

This skill appears to implement a legitimate ZenMux image-generation client, but exercise caution: (1) the package metadata does not declare the ZENMUX_API_KEY even though both SKILL.md and the script require it — verify the publisher and why metadata is incomplete; (2) the provided script appears truncated in the bundle listing (file output was cut off), so request or review the full script before running — the missing tail could contain unexpected behavior; (3) follow the SKILL.md advice and do not pass API keys on the command line; set ZENMUX_API_KEY in a secure environment variable instead; (4) confirm network endpoints (the script calls only zenmux.ai) and, if you must run it, do so in an isolated environment/container and inspect the full source; (5) if you need higher assurance, ask the publisher for complete source and a signed release or prefer an official client from zenmux.ai.

Type: OpenClaw Skill Name: zenmux-image-gen Version: 1.0.0 The skill bundle is primarily benign, providing a legitimate tool for image generation via the ZenMux API. The `SKILL.md` contains clear, non-malicious instructions for the AI agent. The `scripts/generate_image.py` script uses safe practices like `argparse` for argument handling, `urllib.request` for API calls, and `os`/`pathlib` for file operations, mitigating common shell injection risks. However, a vulnerability exists in `scripts/generate_image.py` where the user-controlled `--model` argument is directly inserted into the URL path (e.g., `https://zenmux.ai/api/vertex-ai/v1/publishers/{provider}/models/{model_name}:generateContent`). This could allow for URL path traversal (e.g., `google/gemini-model/../../evil-path`) within the `zenmux.ai` domain, potentially leading to malformed requests or attempts to access unintended API endpoints on the ZenMux server. While this is a vulnerability, it does not demonstrate clear intent for data exfiltration, remote code execution, or persistence on the agent's system, thus classifying it as 'suspicious' rather than 'malicious'.