refactor

Pass

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted inputs (the legacy code and refactor goals) without using security boundaries or delimiters to isolate user data from the instructions.
  • Ingestion points: Untrusted content enters the agent's context through the 'Code' and 'Refactor goal' inputs defined in SKILL.md.
  • Boundary markers: There are no markers (such as XML tags or triple backticks) suggested to encapsulate user data, which increases the risk that malicious code could hijack the prompt's intent.
  • Capability inventory: The skill does not restrict tool access in its frontmatter configuration; therefore, it may inherit high-privilege capabilities from the host environment, such as file system or network access.
  • Sanitization: The instructions do not include any steps to validate or sanitize the input provided by the user.
Audit Metadata
Risk Level
SAFE
Analyzed
May 7, 2026, 09:40 PM
Security Audit — agent-trust-hub — refactor