Optimizes Claude Code sessions by managing token budgets through strategic navigation and subagent isolation.
Context Discipline provides a rigorous framework for managing Claude Code's 200k token window, treating it as a finite resource rather than an unlimited dumping ground. It equips Claude with strategies like MCP-first navigation, lazy loading, and subagent offloading to ensure peak performance during long-running sessions and complex tasks. By prioritizing summarized insights over raw file reads, this skill prevents session sluggishness and keeps Claude focused on the most relevant information in large-scale codebases.
Key Features
01Provides context pruning and summarization strategies for long-running development sessions
02Implements subagent isolation to offload heavy exploration and keep the main session clean
03Prioritizes MCP-first navigation to reduce token usage by up to 9x compared to file reads
04Includes a decision matrix for efficient file reading and budget planning
05Enforces lazy loading patterns to prevent upfront context saturation
06189 GitHub stars
Use Cases
01Maintaining Claude's accuracy and responsiveness during long, multi-turn coding sessions
02Navigating enterprise-scale codebases with hundreds of projects without hitting token limits
03Performing deep-dive research and cross-cutting analysis using isolated subagents
