context-budget-plus

Built on affaan-m/ECC by @affaan-m (249,324 stars, MIT). All credit for the original idea to them. This fork improves and repackages it; upstream license preserved in UPSTREAM_LICENSE.

Find what fills your Claude Code context, spot waste, and rank safe ways to save tokens.

Built for Claude Code users with many skills, agents, rules, MCP servers, or CLAUDE.md files. It plugs into your current setup and only reads your files and settings.

🔍 Why

A large setup can fill context before your real work starts.

Old rules may still load. Skills may be copied more than once. MCP servers may add many tool schemas. Long CLAUDE.md files may repeat the same advice.

This skill checks each part, shows the facts it can prove, and keeps rough guesses clear.

🚀 Install

Run one command:

mkdir -p ~/.claude/skills/context-budget && curl -fsSL https://raw.githubusercontent.com/context-budget-plus/context-budget-plus/main/skill/SKILL.md -o ~/.claude/skills/context-budget/SKILL.md

The skill is a single Markdown file. It has no added packages.

🧭 Usage

Ask Claude Code:

/context-budget

For the full file and tool list:

/context-budget --verbose

Expected output:

Context Budget Report
=======================================

Scope: current project and user settings
Model window: unknown
Estimate method: word count × 1.3 and character count ÷ 4
Total estimated overhead: ~[estimate] tokens
Estimated available context: unknown
Missing or unreadable sources: [count]

Component Breakdown:
| Component | Count | Estimated tokens |
|---|---:|---:|
| Agents | [count] | ~[estimate] |
| Skills | [count] | ~[estimate] |
| Rules | [count] | ~[estimate] |
| MCP tools | [count] | ~[estimate] |
| CLAUDE.md files | [count] | ~[estimate] |

Top 3 Changes:
1. [safe change and likely savings]
2. [safe change and likely savings]
3. [safe change and likely savings]

The skill follows four simple rules:

1. Read only. Do not change files.
2. Measure facts when possible.
3. Label rough guesses.
4. Do not advise removal when use is unknown.

🛠️ What we changed vs upstream

• Rewritten from Japanese into clearer, simpler English with a shorter description and more scannable instructions.
• Adds explicit read-only safety guidance, graceful handling of unreadable paths, and an “Unknown” category to prevent unsupported removal advice.
• Makes duplicate detection more rigorous by using hashes for exact copies, requiring evidence for near-duplicates, and checking symlinks/generated files.
• Qualifies MCP estimates and removal recommendations, distinguishing measured facts from rough guesses and considering benefits such as authentication or structured data.
• The provided improved version expands audit edge cases but omits the original’s report format, verbose-output specification, examples, and best-practices section.

📄 License

Released under the MIT License.

The upstream MIT license and credit are preserved in UPSTREAM_LICENSE.