Add: skill-optimizer — Static LLM tool-calling benchmark and doc optimizer

[damienen]

skill-optimizer benchmarks whether LLMs call the right SDK, CLI, or MCP tools from your guidance docs (SKILL.md), using static action + argument matching. It also runs an iterative optimizer that rewrites your docs until every configured model meets a per-model score floor — producing CI-ready PASS/FAIL verdicts.

Fits under "Core Frameworks" alongside Promptfoo and Inspect AI as a static, offline evaluation harness with a focus on tool-calling correctness rather than answer quality.

Repo: https://github.com/fastxyz/skill-optimizer

Checklist:

• Entry is in scope for this list (static LLM evaluation harness)
• Alphabetically placed (S, after Ragas, before TruLens)
• No duplicates
• Badge appears before the link, matching existing format