Skill

audit-context

Analyze your current context token overhead. Find waste from bloated CLAUDE.md, unnecessary skills, and idle MCP servers. Use to optimize your setup.

Install

npx claudepluginhub artmin96/forge-studio --plugin context-engine

Tool Access

This skill is limited to using the following tools:

ReadBashGlobGrep

SKILL.md

Similar Skills

cache-components

Guides Next.js Cache Components and Partial Prerendering (PPR) with cacheComponents enabled. Implements 'use cache', cacheLife(), cacheTag(), revalidateTag(), static/dynamic optimization, and cache debugging.

cache-components

138.8k

claude-opus-4-5-migration

2 files

Migrates code, prompts, and API calls from Claude Sonnet 4.0/4.5 or Opus 4.1 to Opus 4.5, updating model strings on Anthropic, AWS, GCP, Azure platforms.

claude-opus-4-5-migration

83.2k

bmad-help

Analyzes BMad project state from catalog CSV, configs, artifacts, and query to recommend next skills or answer questions. Useful for help requests, 'what next', or starting BMad.

bmad-pro-skills

43.8k

Stats

Parent Repo Stars1

Parent Repo Forks0

Last CommitApr 3, 2026

Actions

View Source View Plugin View on GitHub View README

Step 1: Measure CLAUDE.md Weight

Read all CLAUDE.md files that would load for this project:

~/.claude/CLAUDE.md (global)

./CLAUDE.md (project root)

Any parent/child directory CLAUDE.md files

For each: count lines, estimate tokens (~1.3 tokens per word), flag if over 100 lines.

Ceiling: ~150-200 instructions max before compliance drops. Claude's system prompt uses ~50. That leaves 100-150 for you.

Output

CONTEXT AUDIT ============= CLAUDE.md: [X lines / ~Y tokens] — [OK / OVER BUDGET] Plugins: [X enabled] — [List any irrelevant to current project] MCP Servers: [X active / ~Y tools] — [OK / TOO MANY] Auto-Skills: [X loading descriptions] — [List any that should be disable-model-invocation] TOTAL ESTIMATED OVERHEAD: ~[N] tokens per session RECOMMENDATIONS: - [Actionable item 1] - [Actionable item 2]

Audit Context: Find and Fix Token Waste

Context tokens are currency. Everything loaded competes with your actual work. This audit identifies waste.

Step 1: Measure CLAUDE.md Weight

Read all CLAUDE.md files that would load for this project:

~/.claude/CLAUDE.md (global)
./CLAUDE.md (project root)
Any parent/child directory CLAUDE.md files

For each: count lines, estimate tokens (~1.3 tokens per word), flag if over 100 lines.

Ceiling: ~150-200 instructions max before compliance drops. Claude's system prompt uses ~50. That leaves 100-150 for you.

Step 2: Check Installed Plugins

Run: cat ~/.claude/plugins/installed_plugins.json (or check settings.json enabledPlugins)

For each enabled plugin: is it relevant to this project? If not, it's loading skill descriptions into context for nothing.

Step 3: Check MCP Servers

Look at .mcp.json files and settings for active MCP servers. Each loads tool names at minimum.

Rule of thumb: Under 10 MCP servers, under 30 tools total. More = context bloat.

Step 4: Check Skills

List all skills that auto-load (those without disable-model-invocation: true). Each one costs description tokens every session.

Output

CONTEXT AUDIT
=============
CLAUDE.md:    [X lines / ~Y tokens] — [OK / OVER BUDGET]
Plugins:      [X enabled] — [List any irrelevant to current project]
MCP Servers:  [X active / ~Y tools] — [OK / TOO MANY]
Auto-Skills:  [X loading descriptions] — [List any that should be disable-model-invocation]

TOTAL ESTIMATED OVERHEAD: ~[N] tokens per session
RECOMMENDATIONS:
- [Actionable item 1]
- [Actionable item 2]

Key Insights

A focused 30-line CLAUDE.md outperforms a 200-line one
Skills with disable-model-invocation: true cost ZERO until invoked
MCP tool schemas load on-demand by default (good — keep it that way)
Plugins you don't use in this project should be disabled per-project