Misha Kaletsky

expect-type lets TypeScript projects test public and internal types with an API that feels close to expect, except the failure happens in the compiler. The machine refuses the shape before runtime ever gets a vote.

It has no meaningful runtime behavior. That is the trick. Type regressions do not need a second testing universe.

trpc-cli turns an existing tRPC or oRPC router into a CLI with typed inputs, validation, help text, subcommands, prompts, completions, and a small dependency surface.

The useful trick is reusing the router contract. A procedure that already works as application code becomes a command without a second parser and a second little pile of lies.

sqlfu starts from the premise that the data layer should be reviewable as SQL. Schemas live in definitions.sql, migration history lives in SQL files, and queries live in .sql files. TypeScript is generated around the evidence.

The current shape includes schema diffs, migration drafting, typed query wrappers, SQLite-compatible adapters, formatting, linting, tracing, and project tooling. Pre-alpha, but stubborn in the right places.

fetchomatic is a small work-in-progress wrapper around fetch: retries, timeout, logging, caching, error handling, and other suspiciously necessary conveniences.

It is not trying to become a religion. The ambition is lower and more useful: make the boring request path observable, survivable, and slightly less eager to ruin an afternoon.

mopub is a tiny publishing tool for pnpm monorepos. It handles the repetitive package-release machinery I do not want to think about every time another small tool needs to leave the house.

It is not glamorous. That is a feature. The publishing tool should be the boring part.

schematch is schema-first pattern matching for TypeScript. Give it zod, valibot, arktype, or anything else that speaks Standard Schema, and use those validators as match cases.

I forgot I made this. I still think the shape is right. It belongs somewhere lower down in the runtime, possibly Bun, if anyone is feeling unusually reasonable.

artifact.ci is a GitHub App that notices uploaded GitHub Actions artifacts and gives them a browsable web URL. It lives in the tedious middle ground where CI already produced something useful, like a Playwright report, coverage report, static site, PDF, or image, but GitHub still wants you to download a zip and rummage through it.

Install the app, keep using actions/upload-artifact, and the useful thing gets a door. Not a platform. A door.

eslint-plugin-codegen uses ESLint as a place to keep generated code and generated documentation synchronized. A comment marks the generated block, a preset defines how to produce it, and normal lint or fix-on-save workflows keep it current.

Useful when the same facts must appear in multiple places and everyone pretends they will remember to update both. Barrels, markdown from JSDoc, tables of contents, labels, examples from tests. Little recurring debts.

pgkit is a PostgreSQL and TypeScript toolkit rather than an ORM. Typed SQL client, query generation, schema inspection, migrations, diffs, admin UI. A lot of machinery from the period when the problem was still called Postgres.

The useful parts are moving to sqlfu. The rest can sit there, making a faint electrical noise.

Umzug is a migration runner for Node projects. It gives you a programmatic API and CLI for running, rolling back, listing, and storing migrations without tying the idea of a migration to one specific database framework.

I do not really use it anymore, but those Sequelize people do. Fair enough. Old tools are allowed to keep working.