What problem does it solve? Long AI coding sessions degrade as the context window fills up, causing repeated answers, forgotten instructions, and wasted token budget. This Skill provides a structured playbook for diagnosing context bloat and applying the right optimization before quality drops. ## Core Features & Use Cases - Context Diagnosis: Threshold-based checks (70% and 90% usage) that tell you when to compact or restart a session. - Optimization Strategies: Immediate actions like /compact, disabling unused MCP servers, delegating heavy output to subagents, and configuring auto-compaction via CLAUDE_AUTOCOMPACT_PCT_OVERRIDE. - Token Efficiency Rules: Output reduction guidelines, prompt scoping techniques, tool-call budgets, and CLAUDE.md size limits. - Use Case: Midway through a large refactor, responses become generic and the session feels slow. Run the diagnosis steps, compact at the task boundary, delegate test output to a subagent, and continue with a clean context. ## Quick Start Ask the assistant to diagnose current context usage and apply the context-optimizer strategies to reduce token consumption in this session.
