Menu Scout
Researches menu gaps and proposes new dishes.
It cannot publish. Candidates land hidden and tagged until a person promotes them.
- Menu rules live in the MCP, not the prompt.
- Slack is the review surface.
My stack & tools
Code is the source of truth now. The tools that stay are the ones I can connect, script, and operate from code. Past a point, using the tools stopped being the interesting part. Now I build on top of them: agents, skills, MCP servers, and shared libraries.
Highest leverage
The useful part is not that a model answers. It is what starts it, what it can touch, and where a person still makes the call.
Researches menu gaps and proposes new dishes.
It cannot publish. Candidates land hidden and tagged until a person promotes them.
An iMessage client for the Valet backend.
The prompt is not a security boundary. Product reads and writes run as the user.
Internal agents for sales, content, and job-search loops.
They handle repeatable legwork. A person still decides what gets sent or pursued.
Model work
The interesting problems are permissions, validation, routing, and review.
Built for the stack
These are the reusable parts: skill kits, libraries, MCP servers, and public repos that turn repeated work into systems.
Six Claude Code skills for React Native and Expo apps.
The fix is not a style guide. It is an enforcement system.
Eight skills and a router for shipping marketing sites.
Each skill owns one decision and passes one named artifact downstream.
A TypeScript library for turning taste signals into a profile.
Product logic became a reusable primitive instead of copied app code.
An MCP server for creating dishes under menu rules.
It is the interface agents, Claude Code, and scripts call.
Supporting evidence
The tool list matters less than the interfaces around it. These are the ones that keep earning a place.
Preview deployments are part of the review loop.
Types catch what a confident diff can hide.
How I work
The skills I write are the ones that fix a mistake I watched a model make more than twice.
/fable-briefFable briefOne brief: inputs, architecture, build order, acceptance checks, and verification.
/brand-analystBrand analystFiles decisions, not adjectives, into a searchable library.
/mobile-auditMobile auditCatches raw tokens, missing states, odd spacing, and broken patterns.
/direction-proofDirection proof setKeeps visual direction tied to receipts instead of moodboard language.
/code-reviewMulti-agent code reviewCatches what one pass misses.
Keeps product and distribution in the same loop.
The operating layer for turning vague work into shipped work.
What shaped the thinking
A short shelf of things that changed how I design, ship, or reason about systems.
Sharp design conversations with a guest list worth studying.
Where I go to study type in the wild, especially older references.