Guide · Agent skill
How to use the Context Budget Auditor agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The context-budget skill is a Claude Code skill that audits how much of the model's context window your setup consumes before any work begins. It inventories your agents, skills, MCP servers, rules and CLAUDE.md files, estimates their token cost, and produces a ranked list of cuts. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and how to read its report.
Context Budget Auditoraffaan-m/ECC · MIT Download the skill Quick start
- 1Download the ZIP. Get the context-budget package from this page. The package contains SKILL.md.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill there instead.
- 4Run the audit. Type /context-budget in Claude Code, or ask in plain language whether you have room for more components.
- 5Review the ranked report. Check the component breakdown and the top three optimizations, starting with MCP servers.
- 6Apply the cuts and re-audit. Remove unused or CLI-replaceable servers, shorten heavy files, then run the audit again to confirm the savings.
What the context-budget skill does
Each Claude Code session loads instructions, tool descriptions and project notes into the model's working memory, which is the context window. That space is limited, and every loaded component takes part of it.
When invoked, the agent scans the components in your Claude Code setup, estimates the token overhead of each one using rough word-count formulas, flags bloated or duplicated items, and writes a report with suggested savings such as removing unused MCP servers or lazy-loading rules.
- Agents: files in agents/*.md, flagged if over 200 lines or if the description frontmatter exceeds 30 words
- Skills: skills/*/SKILL.md, flagged if over 400 lines; duplicate copies in .agents/skills/ are skipped to avoid double counting
- Rules: rules/**/*.md, flagged if over 100 lines; overlapping content between rule files is detected
- MCP servers: .mcp.json or the active MCP configuration, with an estimate of about 500 tokens per tool schema
- CLAUDE.md files: project and user level, flagged if the total exceeds 300 lines
Install the skill from the ZIP or the Claude apps
Option one: download the ZIP from this page and unzip it into your user skills folder, ~/.claude/skills/, so the skill is available in every project. To limit it to one project, unzip it into that project's .claude/skills/ folder instead.
Option two: upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Files in the package: SKILL.md only
- The skill runs no scripts and makes no network calls, according to the item details
Trigger the audit with natural requests
You can run the audit with the slash command /context-budget, or ask in plain language. The source says the skill supports the /context-budget command. Add --verbose to get the detailed mode.
Good triggers include asking whether you have room for more components, or reporting that sessions feel slower or less accurate after adding skills, agents or MCP servers.
- /context-budget
- /context-budget --verbose
- I want to add 5 more MCP servers, do I have room?
- My sessions have become slow since I added a lot of skills. Can you check my context budget?
/context-budget --verboseRead the report and act on the top savings
The report lists estimated overhead per component type, the effective available context as a percentage, a list of issues ranked by token savings, and the top three optimizations. The source example assumes a 200K-token window for Claude Sonnet.
Components fall into three buckets: always needed (keep), sometimes needed (consider on-demand activation), and rarely needed (remove or lazy-load). Detailed mode adds per-file token counts, line breakdowns of heavy files, duplicated lines side by side, and per-tool MCP schema sizes.
- Start with MCP servers: each tool schema is roughly 500 tokens, so a server with 30 tools costs more than all of your skills combined, according to the source
- Remove or replace servers that only wrap simple CLI commands such as gh, git, npm, supabase or vercel
- Shorten agent descriptions over 30 words, since they load on every Task tool call
- Run the audit again after adding agents, skills or MCP servers to catch growth early
Understand the estimates and their limits
The skill estimates tokens with words × 1.3 for prose and characters ÷ 4 for code-heavy files. These are approximations, not exact counts from the model's tokenizer, so treat the figures as a guide for comparing components rather than precise measurements.
- Use the numbers to rank what to cut, not to predict exact limits
- Use --verbose only when you need to pin down which file drives overhead; the source says it is not for routine audits
What people use it for
- Checking whether a Claude Code setup with many MCP servers still has room before adding another one
- Finding agent descriptions and rule files that load into every session without being used
- Trimming a shared CLAUDE.md and overlapping rule sets across a team's projects
- Diagnosing why a long-running session has become slower or less accurate after adding components
Frequently asked questions
How do I install the context-budget skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder for one project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Is the context-budget skill free?
The source lists the licence as MIT. It does not mention pricing, so check the upstream repository for any terms beyond the licence.
What does the context-budget skill require?
It requires Claude Code with agents, skills, rules and MCP configuration present in your project or user setup. The source lists only the SKILL.md file and does not specify external libraries, API keys or accounts.
Does the skill run scripts or use the network?
According to the item details, it runs no scripts and makes no network calls. It contains only SKILL.md.
Will it give me exact token counts?
No. It estimates tokens with words × 1.3 or characters ÷ 4, so the figures are approximate and best used to compare components and rank savings.
Should I use this if I only use the Claude apps?
Probably not. The scans target Claude Code's agents, skills, rules and MCP folder structure, so outside that setup the audit may find nothing useful.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use context-budget if you run Claude Code with many skills, agents or MCP servers and want a ranked view of what is consuming your context window. Skip it if you use the Claude apps or other tools without Claude Code's folder structure, or if you need exact token counts.
Download the skill