The Agent Skills Directory

Indirect Prompt Injection (SAFE): The skill processes untrusted data from Douban RSS feeds and HTML pages.\n
Ingestion points: scripts/douban-rss-sync.mjs (RSS feed content), scripts/douban-scraper.mjs (HTML content), and scripts/douban-browser-scraper.mjs (DOM elements).\n
Boundary markers: Not present in the resulting CSV files.\n
Capability inventory: The skill has filesystem write access (fs.writeFileSync, fs.appendFileSync) and performs network requests (https.get, fetch).\n
Sanitization: Implements csvEscape to ensure valid CSV formatting (escaping quotes, commas, and newlines), which mitigates structural data injection but does not filter content for downstream LLM instructions.\n- Dynamic Execution (SAFE): scripts/douban-browser-scraper.mjs uses dynamic code evaluation to bridge logic into a browser context.\n
Evidence: Uses new Function() to wrap scraping logic before passing it to page.evaluate via Puppeteer.\n
Analysis: The script strings (parseScript, parseGameScript) are hardcoded constants within the source code. No user-controlled or external input is used to generate the executed scripts, eliminating the risk of arbitrary code injection.\n- Privilege Escalation & Persistence (SAFE): No attempts to acquire elevated permissions or establish persistence were found. While the documentation suggests setting up a cron job, the skill itself does not modify system configuration or schedule tasks.

douban-sync