skills.rest
findexu
findexuCOMMUNITY
Explore all skills by this author →
GitHub

context-budget

Estimate per-turn token costs from .claude/ configuration and CLAUDE.md.

1|Updated May 27, 2026
Search Tags:#Claude Code#context window#token budgeting#configuration auditing#prompt cost#workflow skills#heuristic vs exact counting
One-click install
npx skills add https://github.com/findexu/finpack-claude --skill context-budget-findexu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: context-budget
Source: https://github.com/findexu/finpack-claude/tree/main/plugins/setup-finpack/template/skills/context-budget
Command: npx skills add https://github.com/findexu/finpack-claude --skill context-budget-findexu

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Context-budget helps you estimate the per-turn token usage caused by your project’s .claude/ configuration and CLAUDE.md, so you can avoid silent prompt bloat and understand what will load every turn versus only when invoked.

Core Features & Use Cases

  • Always-loaded vs scoped vs invoked-only breakdown: Identifies which rules, skills, and agents contribute to every-turn context versus those that only load when paths match or when commands are triggered.
  • Token estimation with two methods: Uses a chars/4 heuristic by default, or supports exact token counting via Anthropic’s count_tokens endpoint when --api is provided and an API key is available.
  • Budget verdict and trimming guidance: Aggregates totals, ranks the top always-loaded contributors, flags overruns, and recommends the highest-leverage trim target.

Quick Start

Run the context-budget skill to generate a token-cost report for your current project’s .claude configuration and CLAUDE.md.

Frequently Asked Questions about context-budget

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I estimate per-turn token cost from Claude Code configuration files?▼

Estimate per-turn token cost from Claude Code configuration by classifying always-loaded, path-scoped, and invoked-only components in your .claude/ directory and CLAUDE.md, then aggregating their token counts to reveal prompt bloat.

What is the difference between always-loaded and invoked-only context in CLAUDE.md?▼

Always-loaded context in CLAUDE.md contributes tokens every turn, while invoked-only context loads solely when commands are triggered. Classifying these components isolates silent prompt bloat from rules loading unnecessarily.

How do I get exact token counts instead of heuristic estimates for prompt budgeting?▼

Get exact token counts for prompt budgeting by running the skill with the --api flag and providing an ANTHROPIC_API_KEY, which utilizes Anthropic's count_tokens endpoint instead of the default chars/4 heuristic estimation.

Why does my Claude Code context window fill up so quickly during workflows?▼

Your Claude Code context window fills quickly when always-loaded rules and skills contribute more tokens than intended. Auditing your .claude/ configuration identifies top contributors and recommends high-leverage trim targets.

Can I audit which rules and skills are consuming the most token budget?▼

You can audit which rules and skills consume the most token budget by generating a report that ranks top always-loaded contributors, flags context window overruns, and provides trimming guidance for your configuration.

Does context budgeting work without an Anthropic API key?▼

Context budgeting works without an Anthropic API key by default, applying a heuristic chars/4 estimation method to calculate token usage for your .claude configuration and CLAUDE.md files.

Related Skills

Explore More Software Engineering Skills→
bbrowningbbrowning
2

Version Controlling Global Configuration

Version-control portable Claude Code configuration files in ~/.claude with a .gitignore template.

Community
Basic
jadeclijadecli
1

context-audit

Audit conversation context usage and prescribe token compaction strategies.

Official
Intermediate
Sake-TeamSake-Team

context-budget

Analyze Claude Code context window usage and identify token bloat.

Official
Advanced
ShipFailShipFail
1

StopTimizer

Count tokens for GPT, Claude, and Gemini using official tokenizers.

Official
Advanced