subagent-driven-development

Pass

Audited by Gen Agent Trust Hub on Aug 8, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides a structured methodology for coordinating sub-agents. It emphasizes isolation (ensuring sub-agents do not inherit session history) and mandatory review gates, which are positive security and quality practices.
  • [PROMPT_INJECTION]: The skill processes implementation plans which are used to generate tasks for sub-agents. This presents a potential surface for indirect prompt injection.
  • Ingestion points: Implementation plans (e.g., docs/opencode-skills/plans/feature-plan.md) are read to extract task definitions.
  • Boundary markers: The skill instructs the agent to create a "fresh sub-agent per task" with "isolated context" and explicitly states they should "never inherit your session's context or history."
  • Capability inventory: Sub-agents have implementation capabilities including file system access for code changes and tool execution for testing.
  • Sanitization: Handled through the controller agent's curation of task context and the enforcement of multiple review stages.
  • [EXTERNAL_DOWNLOADS]: The license file references a public repository on GitHub (github.com/obra/superpowers) for attribution. This is a reference to a well-known source code hosting service and does not involve automated downloads or code execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 8, 2026, 11:03 AM
Security Audit — agent-trust-hub — subagent-driven-development