deep-research

Pass

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingest untrusted data from multiple external sources including WebSearch results, web_reader captures, and alpha CLI outputs. While the skill employs a 'Reviewer' agent to detect biases and errors, it lacks explicit instruction-isolation boundaries to prevent malicious commands embedded in external content from influencing the agent's logic during the drafting or review phases.
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to interact with the alpha CLI utility (e.g., alpha get <ID>, alpha code <url>). These commands are used to perform academic research and analyze code repositories. Execution of shell commands based on IDs or URLs extracted from search results represents a standard but notable capability.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation (references/tool-map.md) specifies a dependency on the @companion-ai/alpha-hub NPM package, which must be installed globally to enable academic paper analysis features.
Audit Metadata
Risk Level
SAFE
Analyzed
May 8, 2026, 04:00 AM
Security Audit — agent-trust-hub — deep-research