deep-article-research
Pass
Audited by Gen Agent Trust Hub on Feb 25, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and processes untrusted article content from the web.
- Ingestion points: Raw article text is fetched from external URLs provided by the user via the
agent-browserorWebFetchtools inSKILL.md(Step 1). - Boundary markers: The
prompts/deep-verification-prompt.mdfile uses<argument_extraction>XML-style tags to delimit untrusted content, which provides structural separation between data and instructions. - Capability inventory: The agent has access to
Bash(to run the verification script),Write(to save reports),WebSearch, andWebFetch. - Sanitization: No explicit content sanitization or instruction-filtering is performed on the raw article text before it is analyzed by the LLM.
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to execute a local Python script (scripts/gemini_deep_research.py) to perform its primary research function. The script is bundled with the skill and is invoked with controlled arguments.
Audit Metadata