[Feature Request] Configurable budget for available_skills or smarter prioritization
Resolved 💬 3 comments Opened Dec 1, 2025 by connorholly11 Closed Dec 4, 2025
Problem
When using many Agent Skills, Claude Code truncates the <available_skills> section with a message like "Showing 38 of 45 skills due to token limits". This makes skills undiscoverable - Claude can't recommend or auto-trigger skills it doesn't see.
Analysis
Based on research, the <available_skills> section has a 15,000 character budget. Each skill consumes:
- Description text (~261 chars avg)
- XML overhead (~110 chars for tags, name, location)
- Total: ~371 chars per skill
This means only ~40 skills can be displayed regardless of context window size.
The Math
54 skills × 371 chars = 20,055 chars
Budget: 15,000 chars
Overflow: 5,055 chars
Why This Matters
- Skills are designed for progressive disclosure - metadata loads first (~100 tokens), full content only when triggered
- Truncation defeats discovery - users must know skill names in advance
- Alphabetical truncation is arbitrary - skills starting with 't-z' are always hidden
- 15k limit seems low - in a 200k context window, skills could use 50k+ without issue
Request
- Make budget configurable via
settings.json:
``json``
{
"skills": {
"maxCharacters": 30000
}
}
- Or implement smarter prioritization:
- Recently used skills first
- Project-specific skills prioritized
- User-marked favorites
- Reference: Issue #7336 proposes lazy loading for MCP tools - same principle could apply to skills
Environment
- Claude Code version: 2.0.55
- 54 skills across personal + project locations
- All skills have concise descriptions (avg 261 chars, below Anthropic's 305 char avg)
Workaround
Currently consolidating multiple related skills into umbrella skills to stay under 40, but this reduces specificity of auto-triggering.
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗