Add loop-engineering to Development & Code Tools

[qingqingpi]

Adds loop-engineering under Development & Code Tools (alphabetically, before MCP Builder), linking to the maintained repo.

What it is: a skill for designing, evaluating, diagnosing, and hardening repeated or unattended AI-agent loops — maker-checker systems, queue-processing agents, multi-agent refinement. It routes a task to a GREEN / YELLOW / RED fit verdict before designing anything, organized around verifier fidelity (a loop is only as good as its verifier) and a commit gate that holds every irreversible action outside the loop body.

Why it meets the bar:

• Real use case, not hypothetical: agent-loop design, review, and diagnosis.
• Documented: bilingual (English + 中文) README, with on-demand reference files for hardening and deployment.
• Tested: ships its own reproducible eval (evals/) — a paired control/treatment battery across opus, sonnet, and haiku, with a standalone runner and an honest limitations writeup.
• Safe and portable: standard Agent Skills format (SKILL.md + references), MIT, no SaaS dependency and no calls back to any server.

Repo: https://github.com/qingqingpi/loop-engineering-skill

🤖 Generated with Claude Code