cavecrew

To preserve context during small code edits, delegate surgical-edit tasks to compressed subagents instead of handling them inline. This keeps long sessions manageable by relying on structured, low-token output contracts that pass concise results back to the main thread.

You should delegate code review to subagents when you need rapid diff review across long sessions. Delegating diff-review tasks to compressed reviewer agents prevents context bloat while still verifying the changes made during locate and fix loops.

You can chain these workflows efficiently by delegating locate, fix, and verify loops to specialized subagents. Use structured output contracts and scope limits to pass concise results between the investigator, builder, and reviewer agents without wasting main thread context.

Yes, delegating code-location tasks to subagents requires structured output contracts. These low-token contracts ensure the investigator agent returns concise results that are easy to pass back into the main thread for further engineering tasks.

The best way to manage context for symbol definition fixes is delegating the locate, fix, and verify steps to compressed caveman-style subagents. This guides the full delegation flow for one- or two-file changes while keeping the main session lightweight.

Subagent delegation is designed for small-scope edits, concise investigation, and rapid review, not large-scope changes. You should apply scope limits to the builder agent to ensure only one- or two-file surgical edits are delegated, handling larger tasks inline.