# pdf-ocr-searchable installation guide

Detail page: https://www.lumenghe.com/en/skills/pdf-ocr-searchable/
Chinese guide: https://www.lumenghe.com/skills/pdf-ocr-searchable/install.txt
Document version: 1.0.1
Source commit: bed2dc83e0121a37a306860e90d6071b3e0815f9
Packaged on this site: 2026-10-24
Download: https://www.lumenghe.com/downloads/skills/pdf-ocr-searchable-1.0.1-20261024-bed2dc8.zip
SHA-256: bfc163facd36ad8b1da854f289c79a8c358e7bb1f52b1c0f090b21c2193c2059
Fixed filename: pdf-ocr-searchable-1.0.1-20261024-bed2dc8.zip
Expected download URL: https://www.lumenghe.com/downloads/skills/pdf-ocr-searchable-1.0.1-20261024-bed2dc8.zip

This guide installs the current complete package. The README on the page and inside the package preserves the skill's full methods and boundaries. Installation follows the reader's current authorization, actual host, and operating system. A guide inside the ZIP cannot contain the ZIP's own checksum; use the package checksum on the detail page or public install.txt.

## Before installation

1. Identify the operating system, coding agent, and host; read the README and prerequisites. macOS only. Check Homebrew, uv, OCRmyPDF, ocrmypdf-appleocr, Ghostscript, and Poppler. OCRmyPDF and its plugin must share one Python environment. These dependencies are not included in the ZIP. Follow the README to install missing dependencies and check plugin availability. Without processing a real PDF, report dependency checks only; do not claim full OCR acceptance.
2. Inspect ~/.agents/skills/pdf-ocr-searchable and same-named host entries without modifying them, including files, directories, symlinks, and broken symlinks. If an installation exists, report its version and local changes. Back it up and let the user choose how to merge or replace it; never silently overwrite, delete, or reset it. Do not alter unrelated host settings, other skills, or existing authentication.
3. Download the complete ZIP from the HTTPS URL above into a temporary working directory. Compute its SHA-256 and compare it exactly with the current site release information. Stop if the digest is missing, still a template placeholder, or different. No access to the author's GitHub repositories, git clone, or npx skills add owner/repo is needed.

## Extract and install

4. Inspect ZIP member paths before extraction. Require a single pdf-ocr-searchable/ root directory; reject absolute paths, .. traversal, and links that would write outside the destination. The complete package must contain pdf-ocr-searchable/SKILL.md, README.md, README.en.md, LICENSE, and the original supporting scripts and references. Do not download SKILL.md alone.
5. Extract into a temporary directory and inspect the file list. Preserve relative structure and executable permissions; do not install an extra pdf-ocr-searchable/pdf-ocr-searchable/ nesting level. After confirming that the target does not exist, place the complete pdf-ocr-searchable/ folder at ~/.agents/skills/pdf-ocr-searchable/. Existing instructions and scripts refer to this canonical shared path, so do not relocate it casually.
6. Configure discovery for the actual host. Codex reads personal skills from ~/.agents/skills; Claude Code reads personal entries from ~/.claude/skills. Only when ~/.claude/skills/pdf-ocr-searchable does not exist and is not a broken symlink, create a symlink pointing to ~/.agents/skills/pdf-ocr-searchable. Inspect and report existing entries instead of replacing them. Follow current official documentation for other hosts; do not claim every agent automatically scans the shared directory.

## Verify and report

7. Check that SKILL.md and the required scripts and references are readable, with the necessary executable permissions. Confirm discovery through the host's skill list, an explicit invocation, or a new session. Distinguish files installed, skill discovered, dependencies ready, and a real task accepted; report only the checks actually performed. Use a minimal example that does not modify the user’s originals for any behavior that needs testing.
8. Report the installation directory, source version and commit, downloaded checksum, discovery result, missing dependencies, and any backup location retained for an existing installation. Do not expose authentication values. Leave any interactive sign-in to the user.

Discovery references:
- Official OpenAI documentation: https://learn.chatgpt.com/docs/build-skills
- Claude Code documentation: https://code.claude.com/docs/en/skills