vision
Skill2 starsChanged 2 months agoMITView on GitHub
图片识别。主通道 = 会话内主模型(GLM-5.3-flash 原生多模态)直接用 Read 工具读图,零 API 依赖;vision.js 仅作 bash 脚本化备用通道。用于论文附图提取参数、CST模型截图验证、结果曲线对比、图表渲染质量自查等"看图"场景。触发词:看图、识别这张图、图片里有什么、提取图中参数、对比这两张图、模型截图。
- Reads credentials
24 other projects use a version of this.
Adopted by
Most recently taken up by
- kasperjunge/skillstheir own version, first seen Nov 8, 2026
- simota/agent-skillstheir own version, first seen Oct 23, 2026
- sirius-cc-wu/sirius-skillstheir own version, first seen Oct 21, 2026
- knowttl/bstacktheir own version, first seen Oct 20, 2026
- quocld/hephaestus-agent-skillstheir own version, first seen Oct 20, 2026
- frankieramirez/manatheir own version, first seen Oct 17, 2026
Used with
Projects with it also use plan.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
BlockSkillplan
Draft a TDD, PRD, or lightweight plan in the conversation without committing it.
Also in 3 projects that use this
Other versions24 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 24
- gridaco/grida★ 2.7ksame name
- YougLin-dev/Aha-Loop★ 181same name
- xiincs/claude-code-vision-skill★ 170same name
- simota/agent-skills★ 96same name
- quocld/hephaestus-agent-skills★ 8same name
- kasperjunge/skills★ 7same name
- Neostry/vision-skill★ 7same name
- wangbooth/skills★ 7same name
- sirius-cc-wu/sirius-skills★ 6same name
- BhaskarManam/pstack★ 4same name
14 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository9 other files it gives its agents.
- cst-automationSkill
- cst-em-simSkill
- cst-modeling-guideSkill
- cst-result-reportSkill
- em-innovationSkill
- em-simulation-pipelineSkill
- em-verificationSkill
- paper-formula-falsifySkill
- paper-parserSkill
The file
Files
视觉识别 · 主模型原生多模态(2026-08-27 起)
架构变更(2026-08-27):主对话模型升级为 GLM-5.3-flash(原生多模态),本地图片直接用 Read 工具读,不再经 vision.js/API。GLM-4V 系列弃用(4v-plus 余额不足未续)。
通道选择(按序尝试)
| 优先级 | 场景 | 通道 | 说明 |
| 1 主通道 | 会话内看任何本地图片 | Read 工具直接读 | GLM-5.3-flash 原生多模态,零 API、零余额依赖;实测论文 PNG 的文字/布局/裁切检查可靠(2026-08-27) |
| 2 | 远程图片(arXiv/DOI/网图) | curl -L -o /tmp/img.png "<url>" 后 Read | 下载到本地走主通道;墙外源 curl 失败时用通道 3 |
| 3 备用 | bash 脚本/自动化流程(无法"让模型看图"时),或远程图直读 | node ~/.zcode/skills/vision/vision.js [--url <链接>] "<图片>" "<问题>" | 走开放平台 API,默认模型 glm-5.3-flash,需账户余额;429 = 欠费,会话内改走通道 1/2 |
能力边界
| 我能做的 | 我不能做的 |
| 描述天线几何结构、标注尺寸 | 精确测量像素级尺寸(误差 ~0.1mm 量级) |
| 识别层叠结构、材料标注 | 替代 Read 工具读 PDF 文字层(图页同样可 Read) |
| 读取图中的 S 参数曲线趋势 | 精确数字化曲线数据点(需专用工具如 WebPlotDigitizer) |
| 对比两个模型截图的几何差异 | 判断 0.5mm 以下的尺寸差异 |
| 识别公式截图中的符号和结构 | 解析复杂矩阵/积分公式(建议结合 LaTeX 文本) |
| 检查图表渲染质量(文字裁切/遮挡/字号,2026-08-27 实测有效) | — |
触发场景
以下情况优先用 Read 工具直接读图(主通道);只有 bash 脚本自动化流程才调 vision.js:
- 论文附图:天线结构图、尺寸标注图、层叠示意图、加工实物照片
- 结果曲线:S 参数图、方向图、增益-频率曲线(提取大致趋势和数值范围)
- CST 模型截图:检查模型几何是否正确、端口位置是否对齐
- 对比图:仿真结果 vs 论文结果的并排对比
- 报错截图:CST 错误弹窗、求解器发散信息
- 出图自查:generate_figures.py 生成后 Read 复查文字裁切/遮挡(可配合 PIL crop 放大局部)
vision.js 备用通道用法
# 本地图片 — 默认 glm-5.3-flash(开放平台计费,需余额)
node ~/.zcode/skills/vision/vision.js "图片路径.png" "你的问题"
# 网络图片 — 自动下载转 base64
node ~/.zcode/skills/vision/vision.js --url "https://xxx.com/figure.png" "你的问题"
# 指定模型
node ~/.zcode/skills/vision/vision.js --model <模型名> "图片路径.png" "你的问题"技术参数
- 主通道:会话主模型 GLM-5.3-flash(coding-plan 订阅计费,与开放平台 key 是两条通道)
- vision.js:智谱 BigModel 开放平台(open.bigmodel.cn),默认
glm-5.3-flash,key 见同目录.env - 最大输出:2048 tokens(.env 可调)
- 支持格式:JPG, PNG, GIF, WebP, BMP(API 单图 ≤ 5MB;主通道 Read 无此限制)
论文参数提取专用提示词
根据提取目标不同,使用针对性提示词:
提取天线几何结构
请详细描述这张天线结构图:
1. 天线的类型(微带贴片/偶极子/缝隙/Vivaldi等)
2. 各层结构(从上到下),每层的材料和大致厚度比例
3. 所有标注的尺寸符号(如 W, L, h, g 等)及其在图中对应的物理位置
4. 馈电方式(微带线/同轴探针/缝隙耦合等)及其位置
5. 如果用到了多个介质层,说明每层的相对位置提取尺寸标注
请读取图片中所有标注的尺寸数值:
- 对每个标注,给出对应的变量名(如果有)和数值
- 标注单位(mm/mil/μm)
- 如果标注了公差范围,也请列出
- 如果图片中有表格,逐行读取表格数据读取结果曲线
请分析这张 S 参数 / 方向图 / 增益曲线:
1. X 轴和 Y 轴分别是什么物理量,单位是什么
2. 曲线的大致趋势(单谐振/双谐振/宽带等)
3. 关键数据点的大致数值(最低点频率和深度、-10dB带宽范围、峰值增益等)
4. 如果有多个曲线,分别对应什么条件对比模型与论文
请对比这两张图片:
- 第一张是论文中的天线结构图
- 第二张是 CST 模型截图
逐项对比:天线形状、馈线位置、基板比例、端口位置是否一致。
指出任何明显的差异。质量保障
- 对于关键尺寸,优先采信论文正文中的数值,图片仅用于理解空间关系
- 如果图片标注模糊,在 model_spec.yaml 中标记
info_completeness: yellow并记录存疑 - 曲线图提取的数值只能作为参考范围,精确值以论文正文或表格为准
- 如果图片质量太差无法识别,直接报告"图片无法识别"而不要编造
错误处理
API 429 / 余额不足→ 开放平台 key 欠费;会话内直接改用主模型 Read 读图(主通道本来就不依赖余额),或去 open.bigmodel.cn 充值API 4xx→ 检查 GLM_API_KEY 是否有效图片输入格式/解析错误→ 图片可能损坏或格式不支持,尝试另存为标准 PNG/JPG文件不存在→ 检查路径,远程图确认已下载到本地识图失败→ 尝试转换图片格式(PNG → JPG)、降低分辨率,或用 PIL 裁剪关键区域后重读
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

