@aidenwu0209/paddleocr-skills
[English](./README.md) | [简体中文](./README_cn.md)
PaddleOCR Skills
Standalone mirror of the official PaddleOCR Agent Skills from PaddlePaddle/PaddleOCR.
Synced source: PaddlePaddle/PaddleOCR commit f0d83fafe9598134e5ac46aca62a4738f0eabac0
(2026-05-06), checked against upstream main HEAD 211989f0 (2026-06-26).
Upstream refactored the skills in PR #18090 (2026-06-03) — it removed the bundled
scripts/andreferences/and switched to the officialpaddleocr apiCLI. This mirror intentionally keeps the script-based version (which still works and is offline-friendly viauv), and additionally documents the CLI as an alternative path. See Two Ways to Run the Skills below.
Included Skills
| Skill | Use case | Entry script |
|---|---|---|
paddleocr-text-recognition |
Extract text from images, scans, and PDF files | ocr_caller.py |
paddleocr-doc-parsing |
Parse complex documents into Markdown/structured output | layout_caller.py |
Two Ways to Run the Skills
Each skill can be invoked in two ways. The bundled scripts are the default path in this
mirror; the paddleocr CLI is the upstream-canonical alternative introduced in #18090.
| Scripts (default) | paddleocr CLI (alternative) |
|
|---|---|---|
| Install | Just uv — deps are resolved from PEP 723 inline metadata |
pip install "paddleocr>=3.7.0" |
| Required env vars | Per-skill PADDLEOCR_OCR_API_URL / PADDLEOCR_DOC_PARSING_API_URL + PADDLEOCR_ACCESS_TOKEN |
PADDLEOCR_ACCESS_TOKEN only (the CLI resolves the endpoint internally) |
| Output format | {ok, text, result, error} envelope, auto-saved to a temp file |
{jobId, pages:[...]} printed to stdout |
| Page selection (PDF) | Pre-split with scripts/split_pdf.py |
Native --page_ranges "1-5,10" |
| Best for | Skills runtimes, airgapped / offline-friendly setups, no extra install | Environments that already ship paddleocr, or where you want the upstream flow |
The two paths return different output shapes and read different environment variables — they are not output-compatible. Pick one per workflow.
Detailed usage and per-skill examples live in each SKILL.md:
Requirements
- Python 3.9 or later
uv- Internet access
- PaddleOCR official API credentials from paddleocr.com
The scripts use PEP 723 inline dependency metadata, so there are no separate
requirements.txt files to install.
Configuration
Set the environment variables required by the skill you want to use:
| Skill | Required | Optional |
|---|---|---|
paddleocr-text-recognition |
PADDLEOCR_OCR_API_URL ending with /ocr, PADDLEOCR_ACCESS_TOKEN |
PADDLEOCR_OCR_TIMEOUT |
paddleocr-doc-parsing |
PADDLEOCR_DOC_PARSING_API_URL ending with /layout-parsing, PADDLEOCR_ACCESS_TOKEN |
PADDLEOCR_DOC_PARSING_TIMEOUT |
The
PADDLEOCR_*_API_URLvariables above are only required by the bundled scripts. If you use thepaddleocrCLI instead, onlyPADDLEOCR_ACCESS_TOKENis needed — see the "Alternative: paddleocr CLI" section in each SKILL.md.
Local Usage
Run commands from the corresponding skill directory.
cd skills/paddleocr-text-recognition
uv run scripts/ocr_caller.py --file-path "/path/to/image-or-document.pdf" --pretty
cd skills/paddleocr-doc-parsing
uv run scripts/layout_caller.py --file-path "/path/to/document.pdf" --pretty
Install into AI Apps
From this repository root:
npx skills add ./skills/paddleocr-text-recognition -g -y
npx skills add ./skills/paddleocr-doc-parsing -g -y
Or install through OpenClaw:
clawhub install paddleocr-text-recognition
clawhub install paddleocr-doc-parsing
Documentation
- Official documentation mirror: docs/version3.x/deployment/skills.en.md
- Chinese official documentation mirror: docs/version3.x/deployment/skills.md
- Quick reference: docs/QUICK_REFERENCE.md
- Large file guide: docs/LARGE_FILES.md
License
Apache-2.0. See LICENSE.
Loading...
Select a file to preview
Analyzing security...
Checking scan reports and verification data.
Bill of Materials
Everything this skill can do — files, network, commands, and more.