cavecrew

Code reviews on large codebases preserve context by delegating tasks to subagents that provide compressed output. This framework uses a specialized reviewer agent to audit code changes, preventing context overflow during complex workflows.

Locating symbols without exhausting context is best achieved by delegating the search to an investigator subagent. This subagent performs the code location task and returns a compressed output, preserving the main context for other work.

Surgical code edits in complex workflows are made by delegating the modifications to a builder subagent. This targeted approach executes precise edits while maintaining context efficiency through compressed output.

Subagents for task delegation should be used when working on large codebases where context management is crucial. They provide a decision-making framework to determine when code location, surgical edits, and file reviews outperform vanilla commands.

Task delegation with subagents works for managing extended context by utilizing caveman-style agents that output compressed data. This optimizes context utilization for software engineering workflows requiring task-specific context preservation.

Limitations of using subagents for code management include their restriction to specific delegated tasks like code location, edits, and reviews. They function as a decision guide for context preservation rather than a full automation replacement.