Cut and publish a RunTypes release end-to-end — decide the bump, curate CHANGELOG.md, open the chore(release) PR into main, then promote main into prod with a MERGE-COMMIT pull request, watch every workflow, and fix failures forward via PRs into main. Use whenever the user wants to release, publish, cut/bump a version, promote main to prod, ship to npm, finish or unblock a release, or asks why a release workflow is red — even for just one phase (a bump PR, a promotion PR, a failed gate). The agent drives all PRs and CI watching; the developer reviews and clicks the merges.
Entry point for ALL RunTypes enrichment work — the committed, type-keyed FriendlyText<T> (human labels + rt$errors message templates, per-locale translations) and MockData<T> (realistic sample pools/ranges) maps, and the `ts-runtypes` CLI that maintains them (`describe` / `gen` / `gen --update` / `gen --prune` / `gen --translate` / `check`). Use when scaffolding or filling an enrichment mirror file, writing friendly validation errors or field labels, authoring mock data pools, translating a friendly map, filling `@todo` blanks, working with `rt$label` / `rt$errors` / `rt$default` / `rt$items` / pool keys or the `@rtType` / `@rtOrphan` JSDoc tags, or configuring the tsconfig i18n block (enrichment paths are convention under `genDir`, not configurable). This skill only ROUTES: the authoritative skills live in packages/ts-runtypes/skills/ — read the one you need before authoring.
Drive the RunTypes enrichment workflow — author and maintain the committed, type-keyed FriendlyText<T> (human labels + error messages) and MockData<T> (realistic sample data) for a type. Use when scaffolding or filling a type's enrichment file, when running the `ts-runtypes` CLI (`describe` / `gen` / `gen --update` / `gen --prune` / `check`), when filling `@todo` blanks the compiler left, or when working with the enrichment JSDoc tags (`@rtType`, `@rtIds`, `@rtOrphan`, `@rtOrphanChild`, `@todo`). Covers the mirror directory, the compiler-scaffolds/agent-fills loop, the CLI verbs, and the tsconfig i18n block; the per-family authoring DSLs are the runtypes-friendly-type and runtypes-mock-data skills.
Author and use a `FriendlyText<T>` for a RunTypes type — the committed, type-keyed map of human-readable field LABELS + ERROR-MESSAGE templates. Use when writing or editing friendly validation errors, friendly/human-readable field labels, form-builder labels, or a `*.rt.ts` enrichment sibling; when turning `createGetValidationErrors<T>()` output into readable messages via `createFriendlyText<T>(map).errors(...)`; or when an `rt$errors` / `rt$label` / `$[label]` / `$[val]` placeholder template needs writing. Covers the `{ rt$label, rt$errors, ...children }` node shape (total: both meta keys required on every node), the `$[…]` placeholder DSL, the param-precise error-template keys (the failed-constraint name: `type`, `minLength`, `min`, `max`, `pattern`, …), the exclusive `rt$default` catch-all mode, and where the map lives.
Author and use a `MockData<T>` for a RunTypes type — the committed, type-keyed map of realistic sample-value POOLS / RANGES per field that feed `createMockData<T>()`. Use when generating mock data, sample values, realistic test fixtures, or seed data for a type; when authoring or editing a `*.rt.ts` enrichment sibling's mock map; or when wiring per-field `{ pool }` / `{ min, max }` / `{ rt$items, rt$length }` / `{ rt$optional }` overrides into `createMockData<T>({ data })`. Covers the per-field node shape, that pool/range values are validated against the field's type + format at build time (the MD003 rule), and where the map lives.
End-to-end autonomous setup for **RunTypes**. Installs host deps (podman, Node, pnpm, Go, garble), starts the podman engine, bootstraps the tsgolint + typescript-go submodules + patches, installs workspace deps, builds the Go resolver binary + ts-runtypes-devtools, then smoke-tests the docs website container (curl :3000) and the benchmarks container (vite build inside). Use when setting up / bootstrapping RunTypes, installing podman for it, or verifying the containerized apps are runnable. Supports Linux and macOS; prints a not-ready message on other OSes. Specific to RunTypes - NOT a generic project setup (the rest of the monorepo needs only pnpm).
Guide someone through adding a fuzz / property test to their code, the structured way — you investigate their repo, ask focused questions, and iterate until you both have the rules worth checking and the tools to check them. Decide WHAT must always be true before building any tooling. Use when adding a fuzz test, a property test, or any test that throws lots of random inputs at the code; when generalising one example/unit test into a test that runs over many inputs; when deciding what rules to check (round-trip, do-it-twice, compare-to-a-trusted-source, predicted-change, leave-the-rest-alone, reject-bad-input); when testing code with memory by feeding it a sequence of actions (model-based / sequence testing); or when hunting edge cases your hand-written tests miss. Grounded in this repo's real harness (packages/ts-runtypes/test/fuzz/) with the FriendlyText/MockData sync fuzzer as the worked example.
Reset the workspace to start a fresh task in the same shell session — switches to main, pulls the latest HEAD, asks for a new branch name, creates the branch, optionally deletes the merged old branch, and asks whether to compact context. Use when finishing a task and starting a new one without paying the (expensive) cold-boot cost. Triggers: "task reset", "new task", "next task", "start fresh task", "/task-reset".