web-research

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection: The research workflow involves gathering data from external web sources through subagents, saving this content to local files, and then reading those files for synthesis. This creates a surface where instructions embedded in external web pages could potentially influence the agent's behavior during the synthesis phase.
  • Ingestion points: Data enters the environment via the web_search tool (used by subagents) and is subsequently read from local findings files using the read_file tool.
  • Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions found within the gathered research data.
  • Capability inventory: The skill utilizes tools for file system management (mkdir, write_file, read_file, list_files) and agent orchestration (task).
  • Sanitization: There are no explicit instructions for sanitizing or filtering the content retrieved from the web before it is processed by the agent.
  • File System Interaction: The skill creates research directories and writes multiple files (plans, findings, and reports) to the local disk. While these actions are organized within a dedicated subfolder, they represent active interaction with the local file system based on external research topics.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 08:55 PM
Security Audit — agent-trust-hub — web-research