subagent-driven-development

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (LOW): The skill constructs prompts for subagents by directly interpolating external content (task plans and implementer reports) into its core templates. This creates a surface where malicious instructions embedded within a project plan could be executed by the subagents.
  • Ingestion points: implementer-prompt.md (task text) and spec-reviewer-prompt.md (task requirements and report text) ingest untrusted data into subagent contexts.
  • Boundary markers: Absent. The templates use bracketed placeholders for content injection without using XML tags, triple quotes, or specific delimiters to isolate the untrusted data from the system instructions.
  • Capability inventory: The subagents described in the process have significant capabilities, including file modification, command execution for running tests, and version control (git) operations.
  • Sanitization: None detected. The process relies on the controller agent to provide 'full text' of external plans directly into the subagent prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:34 PM