company-research

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform routine filesystem operations such as creating directories, removing temporary files, and executing local Node.js scripts and the browse CLI.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the browse CLI utility via NPM. This is a vendor-supplied tool necessary for the skill's browser-based search and fetch capabilities.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external websites, which presents a surface for indirect prompt injection. The skill implements mitigation strategies including anti-hallucination rules that instruct the agent to avoid inferring products from design elements and to cap fit scores if content is inaccessible.\n
  • Ingestion points: Web content is fetched and cleaned by scripts/extract_page.mjs before being passed to subagents.\n
  • Boundary markers: The subagent prompt templates in references/workflow.md use clear headings and structured sections to separate instructions from external data.\n
  • Capability inventory: Subagents are limited to the Bash tool for performing searches and writing research reports to the local filesystem.\n
  • Sanitization: scripts/extract_page.mjs strips HTML tags and meta-information, providing plain text content for the model to process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:47 AM
Security Audit — agent-trust-hub — company-research