The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external RSS feeds and presents summaries and titles to the agent for semantic screening based on a topic prompt. This creates a vulnerability where malicious instructions embedded in a journal's RSS title or summary could attempt to influence the agent's relevance marking or bypass filtering logic.
Ingestion points: RSS entries fetched from various publisher domains (e.g., nature.com, science.org) and stored in /tmp/sustainability-candidates.json via the collect-window command.
Boundary markers: The workflow does not specify the use of clear delimiters or explicit instructions to ignore embedded commands when the agent evaluates the fetched candidates.
Capability inventory: The skill possesses network access for fetching feeds, local file writing for JSON candidates, and SQLite database modification capabilities.
Sanitization: The script uses standard parsing libraries but does not specifically filter for prompt injection patterns within the ingested feed metadata.
[EXTERNAL_DOWNLOADS]: The skill fetches RSS feed content from several well-known academic publishing services and repositories.
Sources: Includes domains such as nature.com, science.org, sciencedirect.com, wiley.com, frontiersin.org, and tandfonline.com as defined in the assets/journal.opml file.
Dependencies: The instructions require installing the feedparser library from the official Python Package Index (PyPI), which is a standard industry practice for RSS processing.

sustainability-rss-fetch