Aidenwu0209/dsh-PaddleOCR-Skills ↗★ 1
dsh-paddleocr-skills
PaddleOCR text-recognition and document-parsing skills with native DeepSeek Harness tools and GUI configuration.
AI Analysis
集成 PaddleOCR 的文本提取与文档结构解析能力,提供专属设置界面。适合需要对图片或 PDF 进行高精度文字识别、表格提取和布局分析的用户。需配置 PaddleOCR 接口端点。
Install
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:Aidenwu0209/dsh-PaddleOCR-SkillsREADME
Read the full README ↗dsh-PaddleOCR-Skills
English | 简体中文
A native DeepSeek Harness bundle adapted from PaddleOCR-Skills. It ships two native tools, two skills, and a dedicated Settings → PaddleOCR GUI.
Included
paddleocr_text_recognitionfor image/PDF text extraction.paddleocr_doc_parsingfor layout, tables, Markdown, and document structure.- GUI fields for both endpoints, timeouts,
uv, result storage, and a DSH Credential reference. - A visible, clickable PaddleOCR official website link, plus direct API-token and official-documentation links in the GUI.
- Tokens stored through DSH Credentials and never returned to the browser.
- Real-path workspace containment for local inputs.
- Auditable raw JSON results under
.dsh-paddleocr/results/by default.
Install
Requires Node.js 22.19+, DeepSeek Harness, Python 3.9+, and uv.
npx @deepseek-ai/dsh plugin --profile web add "github:Aidenwu0209/dsh-PaddleOCR-Skills#main"
npx @deepseek-ai/dsh web
Built lib/ artifacts are committed, so GitHub installation does not require dependency build-script approval.
Open Settings → PaddleOCR. The panel at the top links directly to the PaddleOCR website, the API-token page, and the official API documentation. Then enter the full HTTPS endpoints ending in /ocr and /layout-parsing, keep or change the default PADDLEOCR_ACCESS_TOKEN credential reference, enter the token, and save. Either endpoint may be left blank when only one tool is needed.
Data boundary
Remote tool calls send the selected local file bytes (base64) or HTTPS URL to the externally configured PaddleOCR service. Do not use the plugin for data that is not allowed to leave the workspace. OCR and document content is untrusted data and must never be followed as agent instructions.
The local same-origin GUI may set or remove the token but cannot read it. The host resolves the credential per operation and passes it only through an explicit managed-subprocess environment.
Development
pnpm install
pnpm check
python3 -m compileall -q skills
pnpm pack --dry-run
See UPSTREAM.md for provenance. Licensed under Apache-2.0.