context-budget

The context-budget skill analyzes context window consumption within a Claude Code session to help users understand and optimize token usage across loaded components. It audits agents, skills, rules, MCP servers, and CLAUDE.md files to estimate their token overhead and identify sources of unnecessary context bloat. By measuring file sizes, token estimates, duplicated content, and configuration complexity, the skill provides visibility into how much of the available context window is being consumed before actual user interactions occur. This helps prevent degraded output quality, sluggish performance, and reduced context headroom in large or heavily customized Claude Code environments.

The skill operates through multiple analysis phases. It inventories all relevant components, estimates token usage, and flags oversized or redundant files based on predefined thresholds. It classifies components into categories such as always needed, sometimes needed, or rarely needed to support optimization decisions. The skill also detects specific problem patterns including bloated agent descriptions, heavy agents, redundant skills or rules, excessive MCP server usage, and overly verbose CLAUDE.md chains. Finally, it generates a structured context budget report containing token breakdowns, issue summaries, and prioritized recommendations with estimated token savings. In verbose mode, the report can include per-file analysis, overlap detection details, and MCP tool schema estimates.

This skill is useful for developers, AI workflow maintainers, and advanced Claude Code users managing complex environments with many custom extensions. It is especially valuable for teams that rely on multiple agents, MCP integrations, or extensive rule systems and need to maintain efficient context utilization. The skill can also assist users planning to add new components by estimating available context headroom and identifying optimization opportunities before performance issues occur.