The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ocr_images.py uses subprocess.run to execute swift for macOS Vision OCR and tesseract for fallback OCR. The commands are constructed using lists (avoiding shell injection) and reference a local script vision_ocr.swift provided within the skill's own directory. This is standard behavior for facilitating the stated OCR functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from WeChat article URLs and generates Markdown.
Ingestion points: WeChat article HTML content fetched via urllib in scripts/read_wechat_article.py.
Boundary markers: The skill instructs the agent to use explicit markers like '> 本文件由公众号原文整理为 Markdown' and '## 来源与说明'.
Capability inventory: The skill can write files (.md, .json, .html), download images, and execute local OCR tools via subprocess.
Sanitization: Uses HTMLParser for content extraction and includes manual 'visual verification' steps for the agent to ensure OCR accuracy and fact-checking, which helps mitigate potential injection content from the source article.
[SAFE]: The scripts/profile_store.py utility handles local configuration storage for user preferences. It includes a specific security check to block sensitive field names (e.g., 'token', 'secret', 'api_key') from being persisted in the profile, following best practices for credential safety.

lov-read-wechat-article