The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local shell (scripts/download_wechat_article.sh) and Python (scripts/save_wechat_article.py) scripts to orchestrate the scraping and conversion process. These scripts are invoked with arguments like URLs and output directories provided by the agent.
[EXTERNAL_DOWNLOADS]: The shell script automatically performs an npm install for the playwright-core package if it is missing from the local node_modules directory. Additionally, the Python script uses the requests library to fetch article images from external URLs found within the scraped HTML.
[INDIRECT_PROMPT_INJECTION]: The skill processes raw content from external web pages (mp.weixin.qq.com). Maliciously crafted articles could contain instructions designed to trick the AI agent when it later reads, summarizes, or analyzes the generated Markdown file.
Ingestion points: The scripts/extract_wechat_article.mjs script extracts the full #js_content HTML from the target URL.
Boundary markers: The WechatArticleSaver.build_markdown method in scripts/save_wechat_article.py creates a Markdown file but does not wrap the untrusted article content in boundary markers or include warnings to ignore embedded instructions.
Capability inventory: The skill has the ability to execute shell commands, write files to the workspace, and make network requests via Playwright and Python's requests library.
Sanitization: While filenames and metadata are sanitized for filesystem safety, the main article body is converted to Markdown without filtering or sanitizing potential instruction-like text.
[DATA_EXFILTRATION]: The image downloading logic in scripts/save_wechat_article.py fetches content from URLs extracted from the article's HTML. A malicious article could potentially include URLs pointing to sensitive internal network services (SSRF), causing the script to attempt to download and store that data in the local workspace.

wechat-article-md-local