Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
Visual Muse
v1.4.0ComfyUI 图像生成工坊 — 用自然语言描述需求，自动生成高质量 AI 图片。支持 SDXL/Flux 多模型、风格模板自动匹配、批量生成、质量评分。说「画一张图」即可触发。
⭐ 0· 193·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The declared purpose (local ComfyUI image generation) matches the included files and scripts (ComfyUI client, workflow templates, prompt agents). However some files hardcode host.docker.internal:8188 rather than consistently using the declared COMFYUI_API_URL primary env var, and helper scripts (switch-painter-model.sh) call docker exec on openclaw-gateway which touches other platform components—these behaviors go beyond a simple local renderer and are not fully justified by the skill description.
Instruction Scope
Runtime instructions and included scripts perform many privileged actions: cloning repositories, pip installing PyTorch and dependencies, downloading large model files, copying tools into the OpenClaw workspace, and running docker exec / docker restart on openclaw-gateway. SKILL.md and SOUL.md also contain prohibitions (e.g., '禁止修改 workspace') that conflict with the setup script which writes to the workspace, indicating internal inconsistency in what the agent is allowed to do.
Install Mechanism
There is no formal install spec but a comprehensive setup.sh is included which clones GitHub repos and downloads models from HuggingFace (and mirrored hosts). Hosts are recognizable (github.com, huggingface.co, hf-mirror.com, ghfast.top) so not obviously malicious, but the script triggers heavy network activity and large downloads and installs (torch, requirements) without further sandboxing—this is expected for a local ComfyUI installer but high-impact.
Credentials
The skill declares a single required env var (COMFYUI_API_URL), which is appropriate. However several scripts ignore that env var and hardcode http://host.docker.internal:8188; other parts expect particular filesystem layout (/home/node/.openclaw, ~/ai-studio) and Docker containers (openclaw-gateway). The skill does not request API keys, but it references external LLM aggregator services (ofox.ai) and a Telegram bot—these references imply external network interactions though no credentials are requested.
Persistence & Privilege
always:false (good). Nevertheless the included scripts perform persistent changes: copying tools/workflows into the agent's workspace, creating start scripts, downloading models, and scripts that invoke docker exec/docker restart on openclaw-gateway (which can alter running gateway/container state). Modifying other platform components (openclaw-gateway) is significant and not limited solely to the skill's own files.
What to consider before installing
This skill is plausible for local ComfyUI-based image generation, but review and take precautions before running its setup scripts:
- Inspect scripts (setup.sh, paint-dispatch.sh, quick-generate.sh, switch-painter-model.sh) line-by-line. They will clone repos, pip install torch and other packages, and download multi-gigabyte model files.
- Note that some scripts hardcode http://host.docker.internal:8188 instead of using COMFYUI_API_URL; if your ComfyUI runs elsewhere update those invocations first.
- The setup and helper scripts copy files into ~/.openclaw/workspace and create ~/ai-studio, and switch-painter-model.sh runs docker exec / docker restart on the openclaw-gateway container—only run these if you trust the environment and have backups (or run in an isolated VM/container).
- The skill references external hosts (GitHub, HuggingFace and mirrors, ghfast.top, hf-mirror.com, ofox.ai, a Telegram bot). Expect network traffic to those endpoints when installing or when invoking model-switching features.
- If you only want the prompt-generation/runtime behavior, avoid running the full setup; instead run the specific scripts you reviewed or adapt the Python tools to point to your existing ComfyUI instance.
- If unsure, run the installer in a disposable VM/container, or manually perform the steps after verifying each remote URL and command. Do not provide unrelated credentials—no extra API keys are required by the skill itself, but gateway/container operations may rely on credentials held by your system.Like a lobster shell, security has layers — review code before you run it.
latestvk975gwpvya0089tvbyra484aeh839p75
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
🎨 Clawdis
Binspython3, curl
EnvCOMFYUI_API_URL
Primary envCOMFYUI_API_URL