subagent-driven-development

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill architecture involves interpolating untrusted data from implementation plans directly into subagent instructions, creating a vulnerability to indirect prompt injection.
  • Ingestion points: Found in implementer-prompt.md (task text) and spec-reviewer-prompt.md (requirements text) where data from external plans is directly inserted.
  • Boundary markers: While Markdown headers are used for organization, the templates lack explicit instructions or robust delimiters to prevent subagents from obeying commands embedded within the plan data.
  • Capability inventory: Implementer subagents are granted general-purpose tool access, which typically includes filesystem operations and shell execution, to carry out their assigned tasks.
  • Sanitization: There is no evidence that the plan content is validated, sanitized, or escaped before being included in the subagent's execution context.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 04:25 PM