Skills Plugins MCP Prompt Model 博客 我的中心

autocad-image-redraw

Convert raster drawing references—including screenshots, scans, photos, sketches, exported PNG/JPG files, and photographed plans—into structured, editable AutoCAD DWG/DXF drawings with explicit evidence tracking, input preflight, specification validation, output inspection, and visual comparison. Use for dimension-driven redraws, hybrid reconstruction, visual tracing, or geometry-only redraws when the source is an image rather than a trustworthy source DWG.

DeepseekModel 官方收录技能 质量 优秀 · 78 v1.0.0

获取

https://deepseekmodel.com/api/download.php?id=pengxiaoan-autocad-dwg-redraw-skill-skills-autocad-image-redraw-skill-md&format=skill
下载 .skill 标准格式,含 system_prompt 与 model_config,导入任意 Agent 框架即可使用
.skill 文件中 system_prompt 字段的实际内容。
name autocad-image-redraw description Convert raster drawing references—including screenshots, scans, photos, sketches, exported PNG/JPG files, and photographed plans—into structured, editable AutoCAD DWG/DXF drawings with explicit evidence tracking, input preflight, specification validation, output inspection, and visual comparison. Use for dimension-driven redraws, hybrid reconstruction, visual tracing, or geometry-only redraws when the source is an image rather than a trustworthy source DWG. AutoCAD Image Redraw Reconstruct an image as an auditable CAD drawing. Treat the image as evidence, not as ground truth. If a trustworthy source DWG exists, use autocad-dwg-redraw instead. Non-Negotiable Rules Never claim physical accuracy from pixel similarity alone. Resolve geometry in this order: explicit user requirements, readable dimensions, derived constraints, calibrated image measurements, then clearly labeled visual estimates. Track every important value as known , scaled , inferred , or unreadable . Calibrate each view or region independently when perspective or differing view scales are present. Do not silently choose between conflicting dimensions. Record the conflict and return needs_review or blocked . Preserve the original image, specs, reports, previews, and hashes so the result can be reproduced. Choose a Profile strict-dimensioned : engineering deliverable; requires readable units and enough known or calibrated anchors to constrain the geometry. general : balanced default for plans, sketches, screenshots, and mixed-quality references. hybrid : use dimensions where available and calibrated proportions elsewhere; label inferred geometry. visual-trace : reproduce visible linework without asserting real-world scale. geometry-only : retain object geometry and useful construction layers; omit dimensions, leaders, notes, and table text. If the source lacks a unit or scale anchor, choose visual-trace or use unitless . Do not default to millimeters merely because the drawing looks technical. Runtime Requirements Use Windows with AutoCAD and pywin32 for DWG generation/inspection. Preflight and comparison require Pillow, NumPy, and OpenCV; DXF preview requires ezdxf and PyMuPDF: python -m pip install pywin32 pillow numpy opencv-python ezdxf pymupdf End-to-End Workflow 1. Preserve and preflight the source Copy the source into the project, compute a hash, and inspect readability before interpreting it: python scripts\preflight_image_redraw.py --image input.jpg --mode general --output reports\input-preflight.json Review rotation, crop, resolution, contrast, blur, perspective, occlusion, visible units, and available scale anchors. A quality warning is not automatically a blocker; it becomes a blocker when the selected profile requires evidence the image cannot provide. 2. Build a semantic redraw spec Use references/image-redraw-spec.md . Create stable entity IDs, identify each view, record calibration anchors and evidence, and encode constraints rather than relying on raw pixel coordinates. Keep source and CAD coordinate systems separate: source coordinates: image pixels, origin at top-left; CAD coordinates: drawing units, origin chosen for auditability; calibration: a view-specific affine, homography, or dimension-based relationship between them. 3. Validate the spec before drawing python scripts\validate_image_redraw_spec.py --spec redraw-spec.json --profile general --report reports\spec-validation.json Resolve errors, impossible geometry, duplicate IDs, missing required fields, invalid units, and constraint failures. In strict-dimensioned , missing units or calibration evidence must block drawing. 4. Draw in AutoCAD python scripts\draw_image_spec.py --spec redraw-spec.json --output outputs\redraw.dwg For Chinese text: python scripts\draw_image_spec.py --spec redraw-spec.json --output outputs\redraw.dwg --text-style CN_TEXT --text-font "C:\Windows\Fonts\simhei.ttf" For geometry-only output: python scripts\draw_image_spec.py --spec redraw-spec.json --output outputs\redraw-geometry.dwg --geometry-only When multiple AutoCAD instances or the Start page confuse COM, use --attach-document with a readable DWG. Use --dry-run before connecting to AutoCAD. For rough, open-source visual tracing without AutoCAD, use image_to_dxf_open_source.py . For vector PDFs, use pdf_vector_to_dxf.py before raster tracing. Neither path converts uncertain pixels into authoritative dimensions. 5. Inspect the CAD output python scripts\inspect_dwg_output.py --dwg outputs\redraw.dwg --report reports\dwg-inspection.json Verify the file opens, ModelSpace is nonempty, units match the spec, expected layers and entity classes exist, and forbidden annotation classes are absent in geometry-only mode. Export an auditable DXF from AutoCAD, then audit and render it: python scripts\render_dxf_preview.py --dxf outputs\redraw.dxf --png reports\cad-preview.png --pdf reports\cad-preview.pdf 6. Compare against the source Create an anchors JSON containing corresponding source-image and CAD-preview points, then run: python scripts\compare_redraw.py --source input.jpg --cad-preview reports\cad-preview.png --anchors anchors.json --output-dir reports\comparison Inspect the side-by-side image, registered overlay, difference view, edge coverage, and edge-distance statistics. Use the comparison to find omitted walls, shifted holes, incorrect arcs, text collisions, and view-layout errors. Do not use it to validate physical dimensions. 7. Iterate and issue a disposition Fix the highest-impact failure, regenerate, and rerun validation. Finish with one status: pass : selected profile and all required checks passed; pass_with_warnings : deliverable is usable with documented noncritical uncertainty; needs_review : a human decision is required, usually due to conflicting or inferred evidence; blocked : required source evidence is missing; fail : output is corrupt or violates a mandatory check. Delivery Contract Deliver as applicable: editable DWG and audit DXF; redraw spec JSON and calibration/anchor JSON; input preflight, spec validation, and DWG inspection reports; fixed-page PNG/PDF preview; side-by-side, overlay, difference image, and metrics JSON; manifest containing source/output hashes, selected profile, units, assumptions, warnings, and final status. References and Scripts references/image-redraw-spec.md : generalized spec, evidence, views, calibration, constraints, and entities. references/image-redraw-validation.md : acceptance policies, checks, metrics, statuses, and report contract. scripts/preflight_image_redraw.py : source-image quality and evidence gate. scripts/validate_image_redraw_spec.py : semantic and constraint validation. scripts/draw_image_spec.py : JSON-to-DWG AutoCAD renderer. scripts/inspect_dwg_output.py : read-only DWG structure inspection through AutoCAD COM. scripts/render_dxf_preview.py : DXF audit and deterministic fixed-page preview. scripts/compare_redraw.py : registered visual comparison and edge metrics. scripts/image_to_dxf_open_source.py : rough raster-to-DXF baseline. scripts/pdf_vector_to_dxf.py : vector-PDF extraction path.
Agent 识别该技能的关键词,点击任意一个即可复制。

该技能未提供触发词。

下载的 .skill 包内含以下字段。
字段 说明
format格式标识(skill/v1)
skill_id技能唯一 ID
name技能名称
version版本号
description技能描述
category所属分类(数组)
trigger_words触发词列表
tags标签列表
source来源标识
source_url来源链接(本页地址)
exported_at导出时间(每次下载生成)
system_prompt系统提示词正文
model_config模型参数:provider / model / temperature / max_tokens / top_p
examples示例
install_guide各平台导入说明(Coze / Dify / Claude / 自定义框架)
同一份技能可按不同平台格式导出。
.skill 标准格式,含 system_prompt 与 model_config,导入任意 Agent 框架即可使用 下载
.skillpro 增强格式,额外含脚本 / 工具 / 依赖 / 钩子占位 下载
.json 纯 JSON 导出,只含 system_prompt 与模型参数 下载
Coze 带 frontmatter 的 Markdown,Coze 平台导入用 下载
Dify Dify DSL,创建应用后直接导入 下载

每日精选 Skill 推荐,免费送到你邮箱

输入邮箱,每天接收一个精选 AI Agent 技能推荐。完全免费,持续更新。

提交后我们会发送一封确认邮件,点击邮件里的链接才会开始收信。

完全免费,取消任意时间。我们不会发送垃圾邮件。