refactor
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection where untrusted application code can manipulate the behavior of the refactoring subagents.
- Ingestion points:
SKILL.mddefines prompts that interpolate untrusted input via the[TARGET_CODE]placeholder across multiple phases (Phases 1-5). - Boundary markers: Absent. There are no structural boundary markers or protective instructions encapsulating the
[TARGET_CODE]text block to prevent it from escaping context and issuing instructions to the subagents. - Capability inventory: The workflow orchestrates the
krakenagent to perform automated file-system modification/code writes, and thearbiteragent to run test suites and linters, which represents a command execution surface. - Sanitization: Absent. Content passed into
[TARGET_CODE]is not validated, escaped, or pre-analyzed for malicious instruction structures before being fed to the subagents.
Audit Metadata