Frontier Plan OpenCode Executor

Core Principle

The frontier plan is a strong hypothesis, not ground truth.

Execute the intent of the plan step by step, but validate every instruction against the actual repository before editing code. Frontier AI plans frequently fail on: nonexistent file names, invented imports, unavailable libraries, patterns that don't match the repo, missing tests, idealized architecture, and breaking internal APIs.

Never batch large changes. Never skip verification. Never hide uncertainty. Never perform destructive actions without explicit approval.

When to Use

• You receive a detailed, multi-step implementation plan written by a frontier AI
• The plan has 3+ distinct steps with specific file paths and code patterns
• You are using a non-frontier coding model that benefits from execution discipline
• The plan carries risk of conflicting with the actual codebase