当内置 image 工具不可用、但本机配置了 GMNCODE_API_KEY 时，使用本地脚本直连 GMNCODE Responses API 完成图片理解。适用于角色识别、图片描述、风格分析、截图理解等任务。

Do not run or grant secrets for this skill until you or the publisher supply and you review the referenced script (/home/ubuntu/.openclaw/workspace/scripts_gmncode_image.py). Specifically: (1) Inspect the script source to confirm what network endpoints it calls, what data it sends, and whether it logs or transmits your files or environment variables. (2) Confirm why the package metadata omitted GMNCODE_API_KEY and request that the skill declare required env vars and include or link the script source. (3) If you must test, use a limited-scope or disposable GMNCODE_API_KEY and run in an isolated environment. (4) Prefer the official image provider when available and rotate or revoke any key used for testing. If the publisher cannot provide the script source or a trustworthy explanation, treat the skill as untrusted.

Type: OpenClaw Skill Name: local-gmncode-vision Version: 1.0.0 The skill bundle provides instructions for the agent to execute a local Python script (`scripts_gmncode_image.py`) at a hardcoded path to handle image processing. However, the source code for this script is not included in the bundle, meaning the agent is being directed to run an unverified local payload. Additionally, the documentation references a non-existent model version ('gpt-5.4'), which is anomalous. While the stated purpose is a benign fallback for image recognition, the lack of transparency regarding the executed code warrants a suspicious classification.