The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted news content from an external RSS feed (smol.ai) which could contain malicious instructions.
Ingestion points: The fetch_news.py script downloads external content from https://news.smol.ai/rss.xml (SKILL.md, Step 2).
Boundary markers: The prompt template in Step 4 uses {content} as a placeholder but does not provide explicit instructions to the model to ignore any embedded directives within that external content.
Capability inventory: The skill allows script execution (python fetch_news.py), file writing (cat > docs/...), and network operations.
Sanitization: The skill performs simple HTML entity cleaning in fetch_news.py but does not sanitize the text for potentially malicious LLM instructions.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data and generate resources.
RSS Fetching: Downloads news data from https://news.smol.ai/rss.xml.
Image Generation: Sends summarized markdown content via a POST request to https://fireflycard-api.302ai.cn/api/saveImg to generate shareable images (SKILL.md, Step 7).
[COMMAND_EXECUTION]: The skill executes a local Python script to perform data fetching.
Pattern: Runs python plugins/ai-daily/skills/ai-daily/scripts/fetch_news.py to retrieve and parse the RSS feed.
File Operations: Uses mkdir -p docs and cat > docs/{date}.html to save generated output to the local file system.

ai-daily-news