subagent-driven-development

Fail

Audited by Gen Agent Trust Hub on Feb 14, 2026

Risk Level: HIGHPROMPT_INJECTION
Full Analysis
  • PROMPT_INJECTION (HIGH): The skill is susceptible to Indirect Prompt Injection (Category 8) due to the way it handles untrusted plan data and code across multiple subagent templates.\n
  • Ingestion points: Untrusted content enters the agent context in implementer-prompt.md (task text interpolation) and spec-reviewer-prompt.md (task requirements and implementation code interpolation).\n
  • Boundary markers: The skill lacks explicit delimiters or "ignore embedded instructions" warnings when injecting external content into the subagent prompts, making it easy for an attacker to override the subagent's instructions.\n
  • Capability inventory: The subagents are granted high-privilege capabilities, including file modification, test execution, and committing code to the repository.\n
  • Sanitization: There is no evidence of sanitization, validation, or schema enforcement for the input plan text or the resulting implementation code before it is processed by the subagents.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 14, 2026, 04:48 PM