[BUG] /skills command shows full skill tokens instead of metadata tokens

What went wrong?

The /skills command displays misleading token counts for skills. Similar to the issue fixed in #14834 for /context, the /skills command shows the full token count of the complete SKILL.md file content, rather than the actual tokens consumed at startup, which should only include the YAML frontmatter metadata (name and description fields).

Since skills use progressive disclosure—only metadata loads initially, with full content loaded on-demand when triggered—the displayed token counts create confusion about actual context usage.

What should happen?

The /skills command should accurately reflect the progressive disclosure model. Two potential solutions:

1. Show dual columns: Display both metadata tokens and full skill tokens in separate columns:

``
Skill Name Metadata Full Content
──────────────────────────────────────────────
commit ~50 2,500
code-review ~80 4,200
``

1. Show dual values inline: Format like "50 / 2,500 tokens" (metadata loaded / full if activated)

The dual-column approach would be most informative, allowing users to see both:

• Metadata tokens (first column): What's actually consumed at startup
• Full content tokens (second column): What gets loaded when the skill is triggered

Steps to reproduce

1. Create or have skills loaded in Claude Code
2. Run /skills command
3. Observe that token counts shown are for full skill content, not just metadata

Environment

• Version: 2.1.1 (Claude Code)
• OS: macOS 26.2 (Tahoe)
• Terminal: Warp
• Platform: Anthropic API
• Model: Opus

Is this a regression?

No - this appears to be the same underlying issue as #14834, which was fixed for /context but not for /skills.

Related Issues

• #14834 - Same bug in /context command (closed/fixed)