# Openclaw Pdf To Cad > OpenClaw Agent Skill for converting PDF engineering drawings into CAD-oriented delivery packages. Use this skill when an OpenClaw user provides a PDF engineering drawing and wants DXF/CAD output, preview files, a delivery package, and a quality report. Do not use it for JPG/PNG-only conversion promises, STEP, IGES, STL, SLDPRT, SLDASM, SolidWorks, assemblies, or other 3D model conversion tasks. - **Type:** Skill - **Install:** `agentstack add skill-cavanjb-openclaw-pdf-to-cad-agent-skill-openclaw-pdf-to-cad` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [CavanJB](https://agentstack.voostack.com/s/cavanjb) - **Installs:** 0 - **Category:** [Content & Media](https://agentstack.voostack.com/c/content-and-media) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [CavanJB](https://github.com/CavanJB) - **Source:** https://github.com/CavanJB/openclaw-pdf-to-cad-agent-skill/tree/main/skills/openclaw-pdf-to-cad ## Install ```sh agentstack add skill-cavanjb-openclaw-pdf-to-cad-agent-skill-openclaw-pdf-to-cad ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # OpenClaw PDF to CAD This is an OpenClaw Agent Skill. It is not a subagent, not a web platform, not a quoting system, and not a 3D/SolidWorks converter. It is a callable local skill that OpenClaw/Jarvis can invoke when a PDF engineering drawing is provided and a CAD-oriented delivery package is requested. ## Role Act as an automation assistant for PDF engineering drawing conversion. Produce a CAD candidate package and clearly report review risk. Do not replace a human engineer's final judgment. ## Inputs Supported baseline input: - `.pdf` engineering drawings, with best results on vector PDFs. Risky or review-required input: - scanned PDFs - image-based PDFs - mixed vector/raster PDFs - blurry drawings - drawings with missing, cropped, or unreadable dimensions - PDF pages containing embedded JPG/PNG drawing images Unsupported input for this skill: - `.jpg`, `.jpeg`, `.png` as standalone source files - STEP, IGES, STL, SLDPRT, SLDASM, SolidWorks, assemblies - 3D-to-2D drawing generation If a user asks for JPG/PNG/scanned drawing conversion, explain that this skill prioritizes PDF engineering drawings. Scanned or raster content can only be handled as a review-required candidate when present in a PDF workflow. ## Processing Contract 1. Treat OpenClaw as the primary runtime for this skill. 2. Verify that the source file is a PDF before running the script. 3. Run the bundled script from this skill folder. 4. Extract vector geometry and text where available. 5. Use OCR only as fallback for scanned/image text and report OCR usage. 6. Preserve Chinese/CJK text as Unicode where possible. 7. Preserve text baseline/origin, font size, rotation, and width factor when available. 8. Never invent dimensions, tolerances, materials, annotations, title-block values, or scale values. 9. Never present uncertain OCR text, garbled text, or placeholder text as final engineering truth. 10. Never send a raw DXF alone when the delivery folder exists. 11. Return the delivery folder, zip path, recommended CAD file, preview PDF, and quality status. ## Required needs_review Conditions Set or preserve `needs_review` when any of the following are true: - The PDF is scanned, image-based, mixed, blurry, low-confidence, or raster-heavy. - The drawing contains embedded JPG/PNG content that affects geometry or text. - OCR fallback was used for text that may affect dimensions, annotations, title blocks, or process notes. - Extracted text contains `??`, replacement glyphs, missing characters, or other garbled markers. - Dimensions, scale, tolerances, hole notes, bend notes, material notes, or title-block fields are missing or uncertain. - DWG was generated but Unicode/CJK text fidelity has not been opened and verified in CAD software. - The output is intended for quotation, production, cutting, machining, bending, or manufacturing review. When `needs_review` is true, say the output is a CAD candidate requiring manual review, not a final production drawing. ## Command ```bash ./scripts/run_pdf_to_cad.sh /absolute/path/to/drawing.pdf --output-dir /absolute/path/to/outputs ``` Optional fields: ```bash ./scripts/run_pdf_to_cad.sh /absolute/path/to/drawing.pdf \ --output-dir /absolute/path/to/outputs \ --customer-name "Customer" \ --project-name "Project" \ --label "delivery-name" ``` ## Output Package Baseline delivery package: ```text _cad_delivery/ ├── .dxf ├── .dwg # optional, only if converter is configured ├── preview.png ├── preview.pdf ├── quality_report.json └── README.md _cad_delivery.zip ``` The package may be mapped into Lite, Pro, Plus/Studio, or Enterprise offerings outside this public skill. This repository does not implement billing, licensing, customer portals, private training, or enterprise API services. ## Output Interpretation The script prints JSON. Use these fields in the agent response: - `status`: `ok`, `needs_review`, or `error` - `delivery_dir`: folder containing all deliverables - `zip_path`: zipped delivery package - `recommended_cad_file`: prefer DWG only if present and safe to recommend, otherwise DXF - `dxf_path`: baseline DXF path - `dwg_path`: optional DWG path when conversion is configured - `preview_png`: visual preview image - `preview_pdf`: human-readable preview - `quality_report`: detailed machine-readable report - `source_type`: vector, scanned, mixed, or unknown/low-confidence - `cjk_text_count`: number of preserved Chinese/CJK text entities - `garbled_text_count`: number of unresolved text entities that extracted as question marks or replacement glyphs - `ocr_text_count`: number of text entities recovered by OCR fallback - `findings`: limitations and review notes When `dwg_path` exists but `recommended_cad_file` remains DXF, tell the user that the DWG was generated but Unicode/CJK fidelity needs manual CAD verification before customer delivery. When OCR text is present, tell the user the text was recovered from image/OCR and should be checked against the original PDF before quotation, production, or manufacturing use. ## Handoff Message Template ```text PDF-to-CAD conversion finished. Recommended CAD file: Delivery ZIP: Preview PDF: Quality status: Review required: Review notes: ``` If status is `needs_review`, include: ```text This is an automatically generated CAD candidate. Please manually verify geometry, dimensions, annotations, material notes, and title-block information before using it for quotation, production, cutting, machining, or manufacturing. ``` ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [CavanJB](https://github.com/CavanJB) - **Source:** [CavanJB/openclaw-pdf-to-cad-agent-skill](https://github.com/CavanJB/openclaw-pdf-to-cad-agent-skill) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-cavanjb-openclaw-pdf-to-cad-agent-skill-openclaw-pdf-to-cad - Seller: https://agentstack.voostack.com/s/cavanjb - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.