discover-plan-batch

To optimize AI agent token efficiency during codebase refactoring, use a discover, plan, and batch execution protocol to front-load context gathering and minimize redundant operations. This structured workflow ensures file modifications are planned before execution, significantly reducing token consumption.

The best way to structure an AI workflow for systematic feature implementation involves enforcing a three-phase protocol: discovering existing code patterns using search tools, planning necessary file modifications in detail, and batching write operations into a single efficient call to maximize output quality.

You gather code context before executing modifications by guiding the AI agent through a discovery phase using glob, grep, and symbols tools. This ensures the agent retrieves existing structural patterns and necessary information before planning any code changes or running execution batches.

No specific dependencies are required to run batched code execution workflows using this protocol. The Skill operates independently to enforce a structured workflow, utilizing internal scripts and references to guide the agent through discovery, planning, and precision execution phases without external libraries.

Yes, you can batch write operations to reduce redundant agent tasks during code modification. The protocol collects all necessary context during a planning phase and then executes all planned write operations in a single batch, ensuring context is gathered before execution to prevent repeated tool calls.

Your AI agent uses too many tokens when implementing new features because it lacks a structured discovery and planning phase, causing redundant operations. Implementing a discover, plan, and batch execution workflow front-loads context gathering to minimize unnecessary token consumption during code modification.