using-mde-skills

Routes ambiguous MDE engineering requests to one canonical execution owner with S4 verification.

1|Updated Sep 15, 2026
One-click install
npx skills add https://github.com/amoai-tech/mdeai --skill using-mde-skills-amoai-tech
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: using-mde-skills
Source: https://github.com/amoai-tech/mdeai/tree/main/.claude/skills/using-mde-skills
Command: npx skills add https://github.com/amoai-tech/mdeai --skill using-mde-skills-amoai-tech

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? When an engineering request in the MDE AI repository is ambiguous or spans multiple domains, it is unclear which skill or workflow should own the work. This Skill resolves that ambiguity by routing each non-trivial request to exactly one canonical execution owner while enforcing independent verification for high-risk S4 changes. ## Core Features & Use Cases - Single-owner routing: Chooses exactly one owner from canonical workflow skills (tasks, systematic-debugging, research, code-review, task-verifier) or domain/vendor skills (copilotkit, mastra, supabase, gemini, stripe) and stops. - S4 safety enforcement: Flags payments, auth/RLS/tenant-boundary, secrets, and irreversible production changes as S4, requiring independent verification by task-verifier instead of self-certification. - Contract-tested routing: Ships a Python test script and 15 eval cases that validate routing rules, retired-owner rejection, and S4 verifier requirements. - Use Case: A request like "Change Supabase RLS tenant access rules" is routed to the supabase skill, marked S4, and blocked from completion until task-verifier independently confirms the work. ## Quick Start Ask the assistant to route an ambiguous MDE engineering request, such as determining which skill should own an unclear cross-repo implementation task.

Frequently Asked Questions about using-mde-skills

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I route an ambiguous engineering request to the right skill?

Invoke this router only when ownership is genuinely ambiguous and the request is non-trivial. It selects exactly one owner, such as tasks for implementation or systematic-debugging for unknown failures, then stops without planning execution steps.

When should I use a routing skill versus calling a domain skill directly?

Call the canonical domain skill directly when one owner is clearly responsible, such as copilotkit for CopilotKit bugs or stripe for payment changes. Use the router only when no single owner is obvious; simple S0 factual questions need no routing at all.

What is S4 risk and why does it require independent verification?

S4 covers payments, auth/RLS/tenant-boundary changes, secrets, and irreversible production side effects. The execution owner cannot self-certify S4 work; task-verifier must independently confirm completion, and failed verification returns the work for correction.

How is the routing contract tested?

A Python script in scripts/ validates the SKILL.md rules against 15 eval cases in evals/routing-evals.json. It checks owner selection, retired-owner rejection, S4 verifier flags, and required routing phrases, printing a pass count on success.

What are the limitations of this routing approach?

The router chooses one owner and stops; it does not plan implementation, manage parallelization, or handle Linear handoff. Requests referencing retired skills like mde-task-lifecycle are rejected as invalid rather than redirected.