experts

Pass

Audited by Gen Agent Trust Hub on Mar 25, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill is inherently exposed to indirect prompt injection because its primary function requires processing untrusted external data from project files and URLs. This is a known risk factor for auditing tools.
  • Ingestion points: Project content is ingested from the local file system via the [PATH] variable and from external websites via the [URL] variable when browser tools are active.
  • Boundary markers: The current subagent prompt template does not utilize explicit delimiters (e.g., XML tags or triple backticks with 'ignore' instructions) to separate the analyzed project data from the agent's core instructions.
  • Capability inventory: Reviewer agents have access to file-reading and browser-inspection tools. However, the skill implements a strict 'no-code guard' ('Do NOT write any code — only research and report findings') which prevents agents from executing instructions found within the data.
  • Sanitization: The skill does not currently include instructions for sanitizing or filtering input project data before it is reviewed by the subagent personas.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 25, 2026, 01:24 AM