executing-plans

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill is designed to ingest implementation plans from external files and use them to drive sub-agent behavior. This creates a critical vulnerability surface where an attacker-controlled plan can hijack the agent's logic.
  • Ingestion points: The skill reads tasks and instructions from a plan-file specified at runtime.
  • Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are utilized when processing the plan content.
  • Capability inventory: The skill can spawn sub-agents, create/modify git branches, execute arbitrary test suites, perform network requests via curl, and commit code changes to a repository.
  • Sanitization: No validation or sanitization of the plan file's content is described.
  • Command Execution (LOW): The skill explicitly uses shell commands for development workflows. This is expected behavior but increases the impact of potential injection attacks.
  • Evidence: Instructions include the use of git for branching and committing, as well as curl for manual verification of API endpoints.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:03 PM