メニュー
Run an HLD + LLD design review on a proposed plan BEFORE coding a non-trivial MemOS feature. Use at the start of any feature that touches more than one module, the data model, or a core invariant.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Write a correct Drizzle schema change plus migration and Row-Level Security policy for MemOS. Use when adding or altering any table, index, or RLS policy.
Generate realistic, screenshot-ready seed data for MemOS — a believable org with teams, projects, agents, OKRs, briefs, and a rich graph of facts/learnings/artifacts. Use before demos, screenshots, or recording the README Loom.
Audit that the MemOS core write-gating invariants (evidence gate, non-obvious gate, provenance thread, tenant isolation) are actually enforced and tested. Run before merging any change to the fact/learning/artifact/workflow write path.
Build a polished MemOS dashboard component to the project design system (Next.js App Router, Tailwind, shadcn/ui, Framer Motion). Use for any new UI piece — OKR tree, activity feed, provenance graph, brief authoring, etc.
Add a new MemOS API intent end-to-end (Zod schema, handler, route registration, test, and API docs) in the consistent project pattern. Use whenever adding or modifying an intent like fact.record, workflow.create, brief.fetch.
Capture an Architecture Decision Record for a real design choice in MemOS. Use whenever a non-trivial decision is made (tech choice, schema modeling, isolation strategy, a tradeoff with alternatives).
| name | design-review |
| description | Run an HLD + LLD design review on a proposed plan BEFORE coding a non-trivial MemOS feature. Use at the start of any feature that touches more than one module, the data model, or a core invariant. |
Before building a non-trivial feature, produce a short design review the human can approve. This is where system-design judgment is shown and captured. Keep it tight — one page — but cover both altitudes.
What are we building and why. What's explicitly out of scope.
db-migration skill).scaffold-intent skill).At least one alternative and why you rejected it. This is the senior-engineer signal.
Anything that could bite: scaling, cost (embeddings!), race conditions, migration safety.
What invariant tests prove this works. What's the demo that shows it.
write-adr skill.scaffold-intent / db-migration / frontend-component as appropriate).The human has read the review and approved (or redirected). Do NOT start coding a multi-module feature without this.