Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed image-generation and image-postprocessing workflow, with practical cautions about external image tools, local file writes, and dependency installation.
Before installing, be comfortable with the skill sending prompts to configured image-generation tools or providers, creating timestamped image folders under Downloads, and installing Pillow if needed. Prefer a virtual environment for dependencies and confirm provider credentials, cost, output count, and destination path before large batches.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
.python-version
# Secrets / local config
.env
.env.local
*.local
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
# Secrets / local config
.env
.env.local
*.local
# Generated test artifacts (kept out of repo)
The declared description says this skill bulk-generates Instagram-aesthetic images from a prompt library and sends them in parallel to image-generation skills, then archives the results. The supplied code does none of that. It is a standalone Python script using PIL/Pillow to open an input image, render styled text overlays in either 'ins' or 'cover' mode, and save a single output JPEG. Its main behaviors are typography/layout, font fallback, CJK/symbol handling, auto-placement over blank image regions, and command-line parsing of input/output/title/body arguments. There is no prompt composition library, no batch orchestration, no parallel dispatch, no image synthesis, and no archival directory creation. This is a clear description-behavior mismatch with a materially different primary purpose.
The declared description says this skill bulk-generates Instagram/Xiaohongshu-style images from a prompt library and sends those prompts to image generation tools in parallel, then archives the results. The supplied code does none of that. It is a standalone collage-maker for combining already-existing images into a grid with configurable layout, gap, background color, cell aspect ratio, and optional final resize. While the code references INS/Xiaohongshu styling in comments and defaults, those are only aesthetic defaults for collage formatting, not image generation. This is a material description-behavior mismatch because the primary function and capabilities are different.
The declared description describes a bulk image-generation orchestration tool: it should compose prompts, invoke image-generation skills in parallel, and save generated images into a download archive folder. The actual code does none of that. Instead, it is entirely a test file for add_ins_overlay.py, validating text rendering behavior such as CJK/font routing, line breaking, blank-region placement, cover-mode layout, and stroke/shadow rendering on images. While the theme is loosely related to Instagram-style visuals, the primary purpose and capabilities are materially different. This is therefore a clear description-behavior mismatch.
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
深色木纹桌面上,摊开的书籍占据画面左侧,银色金属笔斜放在书页上。右侧透明玻璃花瓶中插着鲜红色毛茛花,花瓣层叠饱满。旁边放着显示蓝色屏保的平板电脑,白色无线耳机随意摆在一旁。背景是柔和的白色纱帘,自然光透过窗纱洒入。比例3:4。松弛感ins风
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
深色木纹桌面上,摊开的书籍占据画面左侧,银色金属笔斜放在书页上。右侧透明玻璃花瓶中插着鲜红色毛茛花,花瓣层叠饱满。旁边放着显示蓝色屏保的平板电脑,白色无线耳机随意摆在一旁。背景是柔和的白色纱帘,自然光透过窗纱洒入。比例3:4。松弛感ins风
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
深色木纹桌面上,摊开的书籍占据画面左侧,银色金属笔斜放在书页上。右侧透明玻璃花瓶中插着鲜红色毛茛花,花瓣层叠饱满。旁边放着显示蓝色屏保的平板电脑,白色无线耳机随意摆在一旁。背景是柔和的白色纱帘,自然光透过窗纱洒入。比例3:4。松弛感ins风
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
深色木纹桌面上,摊开的书籍占据画面左侧,银色金属笔斜放在书页上。右侧透明玻璃花瓶中插着鲜红色毛茛花,花瓣层叠饱满。旁边放着显示蓝色屏保的平板电脑,白色无线耳机随意摆在一旁。背景是柔和的白色纱帘,自然光透过窗纱洒入。比例3:4。松弛感ins风
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
深色木纹桌面上,摊开的书籍占据画面左侧,银色金属笔斜放在书页上。右侧透明玻璃花瓶中插着鲜红色毛茛花,花瓣层叠饱满。旁边放着显示蓝色屏保的平板电脑,白色无线耳机随意摆在一旁。背景是柔和的白色纱帘,自然光透过窗纱洒入。比例3:4。松弛感ins风
The guidance says skills doing CJK rendering should "always include Arial Unicode MS" and then narrows Linux to "the equivalent" font choice, effectively prescribing a specific locale/font handling approach rather than offering user choice. Under the policy, forcing a language/locale-related preference without opt-in can be a natural-language policy violation.
The activation text is broad enough to trigger on many generic image or marketing requests, which can cause the agent to invoke this skill outside the user's intended scope. In a tool-using environment, overbroad routing increases the chance of unintended filesystem writes, external model calls, or use of side-effecting commands without sufficiently specific user consent.
Multiple requirements hard-code 小红书-oriented dimensions, Chinese text-length rules, and Chinese-language cover/title conventions as defaults, and the examples/instructions assume Chinese copy formatting. This imposes a specific language/locale style without offering a user choice or documenting that the skill is restricted to a China-specific use case.
The skill reveals the existence and path of another local skill (~/.agents/skills/huny-img/SKILL.md) and names fallback tooling, which leaks internal environment structure. In isolation this is limited, but environment and capability enumeration can help an attacker craft follow-on prompts targeting available tools, credentials, or known sub-skill behaviors.
- **内页模式**：1 prompt 1 张，按用户需要的内页数生成；单图默认 3:4 竖图（小红书内页标准），与 cover 视觉风格一致（色温 / 色调 / 质感统一）
- User may also ask for 3x3 or other collage layouts or customize with given images — calculate count accordingly.
3. Iterate the prompt queue and dispatch them **in parallel** to image generation skills in the following priority order:
- **🥇 首选 `huny-img`**（腾讯混元生图 3.0）— 详情见 `~/.agents/skills/huny-img/SKILL.md`。`hunyuan3-text-to-image.py` 或 `hunyuan3-batch.py`（多 prompt 调度）
- **🥈 备选 `mmx`**（MiniMax 多模态 CLI，`mmx image generate`）— 仅当 huny-img 不可用、缺凭证、或 JobNumExceed 限流时
- **🥉 其他模型**（如 `image-01-live` / Midjourney 等）— 最后才考虑
Suspicious Unicode normalization or mixed-script content
The README's only demonstrated text content is Chinese, and it links to a Chinese README variant without indicating that language choice is optional or that the skill is intentionally limited to a Chinese-language audience. This can be read as a default locale/language constraint without user opt-in, which falls under the language-policy check.
The skill instructions and examples in this file are entirely in Chinese, including the generated text examples and usage guidance, which indicates a language-specific constraint. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern unless the locale restriction is clearly documented and justified.
The skill instructs writing outputs to ~/Download/ins-image-{timestamp}/ without an explicit consent or warning step. Silent filesystem side effects are risky because they can clutter user storage, overwrite expectations about where data goes, and normalize unannounced local writes by agent skills.
The document repeatedly prescribes English-language text elements such as handwritten English cards and black-and-white motivational text cards as preferred composition items. This imposes a specific language aesthetic without user opt-in or justification, which can conflict with language-choice policy expectations.
Several prompt templates explicitly include handwritten English cards as part of the scene composition, reinforcing a fixed language requirement in the generated output. Because the file does not offer alternatives or frame English as optional, this is a natural-language locale/language policy concern.
The user-facing CLI description, examples, option help, and runtime messages are all written in Chinese, which imposes a specific language/locale on users without any opt-in or alternative. Under the policy, forcing a specific language without user choice is a natural-language policy concern.
The CLI help at L955-L956 presents --opacity as a mask opacity setting, which implies it affects rendering. But the code explicitly states at L972-L976 that the value is ignored and only prints a notice, while actual opacity remains hardcoded earlier in rendering logic.
No suspicious patterns detected.