well-actually

Pass

Audited by Gen Agent Trust Hub on Mar 25, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting and processing untrusted data from external websites and source code files.
  • Ingestion points: Agents are instructed to visit user-provided URLs and read various local files (source code, templates, CSS) as defined in the 'Code Access Rules' table in 'SKILL.md'.
  • Boundary markers: The prompt templates provided to the sub-agents lack explicit boundary markers or instructions to disregard potential commands embedded within the reviewed content.
  • Capability inventory: The skill utilizes browser MCP tools for network access and maintains file system read permissions.
  • Sanitization: No sanitization, escaping, or validation of retrieved web content or source code is performed before analysis.
  • [DATA_EXFILTRATION]: The skill allows broad access to source code and utilizes network capabilities.
  • The 'HN Commenter' persona is explicitly granted access to read all source code files in the project.
  • The prompt template directs sub-agents to use browser tools to visit external URLs, constituting network activity that could be leveraged for data exposure if sensitive information is handled.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 25, 2026, 01:24 AM
Security Audit — agent-trust-hub — well-actually