deep-research-skills
Pass
Audited by Gen Agent Trust Hub on Apr 18, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill set documents and provides instructions for several autonomous research agents (e.g., GPT Researcher, Local Deep Research, Tongyi DeepResearch, OpenResearcher) that process untrusted data from external sources. This creates a surface for Indirect Prompt Injection where malicious content in search results or academic papers could influence agent behavior during the synthesis phase.
- Ingestion points: Web search results (Bing, Google, Tavily), academic databases (arXiv, PubMed, OpenAlex), and document uploads (PDF, Markdown, LaTeX).
- Boundary markers: None explicitly documented in the methodology; these tools typically aggregate retrieved text directly into synthesis prompts.
- Capability inventory: Tools like Kosmos and AI Scientist have the capability to write and execute code, run experiments, and generate files based on literature synthesis.
- Sanitization: No evidence of sanitization or validation of the retrieved external content before it is processed by the LLM.
- [COMMAND_EXECUTION]: Several guides include explicit shell commands for installation and environment setup (e.g.,
pip install khoj,docker run ...,pip install deepsearcher). While intended as user documentation, an autonomous agent may attempt to execute these commands directly if instructed to 'setup the research environment' or 'run the research tool'.
Audit Metadata