eceee-news-fulltext-fetch

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests untrusted data from an external source which may be processed by downstream agents.
  • Ingestion points: The fetch_html function in scripts/fulltext_fetch.py retrieves HTML content from the eceee.org news archive and individual article pages.
  • Boundary markers: The extracted article text is stored in a SQLite database and can be output as plain text without boundary markers or delimiters to differentiate it from the agent's own instructions.
  • Capability inventory: The skill has network access (via urllib) to fetch external pages and filesystem access (via sqlite3) to create and update a local database file.
  • Sanitization: Although the skill cleans HTML structure and extracts core text using trafilatura or a fallback parser, it does not perform logical sanitization of the resulting text to remove potential prompt injection instructions.
  • [EXTERNAL_DOWNLOADS]: The skill fetches article metadata and body content from https://www.eceee.org/. This activity is central to the skill's stated purpose of building a news corpus and uses standard networking protocols.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 03:56 AM