Dynamic Context Pruning (DCP)

Overview

DCP is a plugin-based context optimization skill that reduces token consumption by pruning conversation segments that are no longer relevant to the active task. Unlike distill (which compresses code via AST analysis), DCP operates on conversational context — removing stale tool outputs, completed task discussions, and resolved debugging traces.

When to Use

Use DCP when:

• Context budget reaches 65%+ (ContextBridge returns action: 'compress')
• Long conversation with many completed sub-tasks
• Before dispatching subagents (to give them lean context)
• Tool outputs from previous steps are no longer needed
• Multiple debugging iterations have been resolved