deep-research
Warn
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation and methodology strongly recommend the installation of a third-party tool called
search-clifrom an unverified Homebrew tap (199-biotechnologies/tap). This tool is described as the primary retrieval mechanism for the research pipeline. Installing binaries and taps from non-official or untrusted sources represents a security risk as the code and its updates are not reviewed by the platform. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large volumes of untrusted data from the web (via multi-provider searches and content scraping). This data is subsequently passed to sub-agents and synthesis scripts. While the methodology contains prose instructions to treat web content as data, the skill lacks robust technical boundaries (such as specific delimiters or sanitization routines) to prevent malicious instructions embedded in external content from influencing the behavior of the agent or its sub-tasks.
- [COMMAND_EXECUTION]: The orchestration engine relies on the execution of multiple shell commands via the
Bashtool, including interactions with the externalsearch-clibinary and various local Python scripts. This extensive use of shell commands to process external data increases the attack surface, particularly when combined with unverified third-party tools.
Audit Metadata