The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scrapes full article content from the internet (via Sogou WeChat) and asks the agent to summarize it. If a scraped article contains instructions disguised as text, the agent might inadvertently follow them.
Ingestion points: scripts/wechat_search.py extracts full_content from external article URLs and saves it to articles.json.
Boundary markers: The instructions in SKILL.md (Step 2) for generating summaries do not specify the use of delimiters (like triple quotes or XML tags) or instructions to ignore embedded commands.
Capability inventory: The skill utilizes shell execution for Python scripts and has file system write access in the workspace to generate PDF reports.
Sanitization: The scraped text is not sanitized or filtered for instructions before being passed to the LLM for summarization.
[COMMAND_EXECUTION]: The skill relies on executing shell commands to run its internal Python scripts (wechat_search.py, wechat_fetch.py, and wechat_pdf.py). While this is standard for its workflow, it involves the agent interacting directly with the system shell to process user-provided keywords.

wechat-article-scraper