[FEATURE] Per-Skill Model Selection for Cost Optimization

Resolved 💬 2 comments Opened Jan 10, 2026 by raulbarreras Closed Jan 10, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Summary:

Add support for model selection in skill frontmatter to enable automatic model switching based on skill complexity, optimizing for both performance and cost.

Use Case:

Optimize Claude Code costs while maintaining quality. Different skills require different levels of intelligence:

  • Complex skills (threat modeling, architectural security review) → Opus 4.5
  • Standard skills (code review, vulnerability analysis) → Sonnet 4.5
  • Simple skills (formatting, linting, basic checks) → Haiku 4.5

Currently, I must either:

  • Use one model for entire session (suboptimal cost or quality)
  • Manually /model switch mid-session (interrupts workflow, error-prone)
  • Exit/restart sessions with different models (destroys context)

Proposed Solution

Extend skill YAML frontmatter with optional model field:

``` yaml
name: architectural-analysis
description: Deep architectural review and security analysis
model: claude-opus-4-5-20251101


``` yaml
name: code-formatting
description: Apply code formatting standards
model: claude-haiku-4-5-20251001

Expected Behavior:

  • When Claude invokes a skill, temporarily switch to specified model
  • After skill execution completes, revert to session default model
  • If model field omitted, use session default (backward compatible)
  • Model switching transparent to user (optional notification in verbose mode)

Benefits:

  • Cost optimization: 60-80% cost reduction through tactical model selection
  • Quality maintenance: Complex tasks still get Opus-level reasoning
  • Developer experience: No manual model management required
  • Team standardization: Skills encode not just "how" but "with what model"

Alternative Solutions

Subagents with model specification - but these run in isolated contexts, defeating the purpose of maintaining conversation continuity for related tasks.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

_No response_

Additional Context

_No response_

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗