grounding-anything
PassAudited by ClawScan on May 1, 2026.
Overview
The skill is aligned with image grounding, but users should verify the model endpoint, proxy settings, and helper code before using it with sensitive images.
Before installing, confirm that the configured GLM model server is trusted for the images you plan to use, that bypassing the proxy is acceptable in your environment, and that any local helper modules referenced by the skill are reviewed and trustworthy.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Images and prompts may be sent to the configured model server, so sensitive images could leave the local working context.
The skill sends image inputs to a configured model API over HTTP; this is central to the stated image-grounding purpose, but it creates a data boundary users should understand.
url = 'http://<host>:<port>/v1/chat/completions' ... response = HttpInterface(url, prompt, images=[image_path], no_think=True)
Use only with a trusted model endpoint, preferably with appropriate transport security and retention/privacy expectations.
Model requests may avoid the usual proxy path, logging, or filtering for that host.
The skill asks to change proxy routing so internal model requests bypass the proxy; this is disclosed and scoped to the model host, but it can affect network controls.
os.environ['NO_PROXY'] = '<model-host>' # 跳过代理 ... 调用内网模型时必须设置 `NO_PROXY` 环境变量
Confirm this is allowed in the user's environment and keep NO_PROXY limited to the intended model host.
Using or reading local model configuration could expose internal endpoint details if handled carelessly.
The skill points to a local agent model configuration path that is not declared in the registry requirements; no credential use is shown, but this is local configuration access users should notice.
模型 API 地址配置在 `/root/.openclaw/agents/main/agent/models.json`
Ensure only the needed endpoint information is used and do not reveal or copy any secrets that may be present in local model configuration files.
The actual behavior would depend on local helper code that was not part of the reviewed artifact.
The instructions rely on helper modules that are not included in this one-file, no-install artifact set; this is not suspicious by itself, but users should verify those helpers if they are present locally.
from interface_http import HttpInterface ... from utils_boxes import parse_bboxes_from_response, visualize_boxes
Use trusted, reviewed, and preferably pinned helper implementations for the HTTP interface and box-visualization utilities.