The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill ingests untrusted data from external RSS feeds, which creates a surface for potential indirect prompt injection where feed content could attempt to manipulate the agent.
Ingestion points: External RSS feeds fetched in scripts/feeds.py using URLs from scripts/lists.py.
Boundary markers: Absent; summaries are provided as plain text within the JSON output without explicit boundary instructions for the agent.
Capability inventory: The skill allows the agent to execute its own Python scripts and has network access to fetch data.
Sanitization: The clean_html function in scripts/feeds.py strips HTML tags and truncates summaries to 500 characters, providing basic mitigation.
Data Exposure & Exfiltration (LOW): The script performs network operations to multiple external domains to fetch RSS content. While these are curated sources, they are not on an internal whitelist.
Unverifiable Dependencies & Remote Code Execution (LOW): The skill depends on the feedparser package. While it is a standard library for RSS, it is an external dependency installed without a specific version pin.

openclaw-feeds