Claude Code Fixed /context Token Display
Are you debugging why your skills aren’t loading? Have you tried the /context command?
I was trying to figure out why my skill wasn’t loading — had a look in context and noticed that some of my skills were using 1000s of tokens. Some quick maths (using claude of course) showed that I was way over the 200k available.
This is why I created the Fit Skills Plugin - to help me figure out what was going on.
What Changed in v2.1.0
v2.1.0 (January 7, 2026) fixed skill token estimates in /context to accurately reflect frontmatter-only loading. (Claude told me this change was documented in the changelog, but… claude cant find it now 🙄)
| Before v2.1.0 | After v2.1.0 |
|---|---|
xlsx: 6.3k tokens | xlsx: 117 tokens |
pdf: 4.4k tokens | pdf: 68 tokens |
spec: 4.4k tokens | spec: 18 tokens |
The old numbers showed the full SKILL.md token burden. The new numbers just show what’s actually loaded at startup into Claude’s context window.
Why Skills Use Two-Phase Loading
Skills use progressive disclosure:
- Startup: Claude scans all skills but only loads YAML frontmatter (name, description). Cost: ~10-100 tokens per skill.
- Invocation: When Claude decides to use a skill, the full content gets injected. Cost: potentially thousands of tokens.
This is why you can have dozens of skills without blowing your context budget. The /context menu now correctly shows phase 1 costs.
Two Separate Limits
So if tokens were not the issue, why was my skill not being loaded?
1. Description Budget (15k chars default)
Does the total of all your skills, across user, project & plugin spaces combined exceed 15,000 chars?
Check: Install and run Fit Skills to find out.
Fix: SLASH_COMMAND_TOOL_CHAR_BUDGET=30000 claude
This is configurable through an ENV variable but not much info online as to the effects of changing this.
2. Token Limit (~794k observed ceiling) (assumed — personal observation)
Total context window tokens from skills, MCP tools, and memory files. Even if you are under the description budget, - skills can be “hidden due to token limits.”
Check: Ask Claude to look for <!-- Showing x of y skills due to token limits --> in its available skills.
Fix: No env var for this — you must remove items (uninstall plugins, disable MCP servers). Otherwise some of your skills will just disappear.
What /context Shows Now
The token count in /context is the frontmatter cost — what each skill contributes at startup. This helps you:
- See which skills are actually loading
To really understand whats going on:
- Identify heavy consumers of the description budget.
- Debug why project skills might be dropped.
Why wasn’t my skill loading?
My skill was not valid!
Although I asked Claude to create my skill, - Claude did not follow the documented guidelines for skill directory structure.
Is it in a proper directory structure (skills/name/SKILL.md, not skills/name.md)?
Subscribe to Skills Weekly, a digest of the hottest plugins & industry news. Only one email a week.
We respect your privacy. Unsubscribe anytime. Privacy Policy
← Back