NYC

fix-parallel

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • PROMPT_INJECTION (LOW): The skill exhibits a surface for indirect prompt injection. It processes untrusted data from the issues arguments and local project files through multiple autonomous sub-agents. An attacker could place malicious instructions in the code being debugged or the issue descriptions to manipulate agent output.
  • Ingestion points: $ARGUMENTS tag and various files analyzed by the debugger and scout-ext sub-agents.
  • Boundary markers: The skill uses basic XML-like tags () for arguments but lacks strict delimiter enforcement for code files being read.
  • Capability inventory: The workflow includes git-manager (write access), fullstack-developer (code modification), and tester (code execution), which are powerful capabilities if a prompt injection succeeds.
  • Sanitization: Relies on manual approval and a code-reviewer sub-agent, which provides a defensive layer but is not a foolproof security control against adversarial input.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:47 PM