rote-skills

Distributable Claude Code skills for rote.

Three primitives, one idea

rote is how an agent learns the way you learned to ride a bike: do it once, and it sticks. Three primitives carry that, and they're really one motion — reach → record → replay.

Primitive	Motion	What it is	The tax it removes
Adapter	Reach	Your agent talks straight to any API, locally — no gateway, no SDK to wire.	No tollbooth between you and tools you already pay for. Your traffic, your box, your bill — once.
Flow	Record → Replay	As you work, rote keeps what worked and what failed, then crystallizes the wins into reusable memory.	No workflow-vendor subscription. A proven flow replays instead of re-reasoning — determinism, not a metered re-run.
Hub	Replay, shared	A flow or adapter that works becomes your team's and community's — like a teacher's lesson.	No registry tax. One person's mastery is everyone's, with no re-derivation and no per-seat toll on shared knowledge.

The throughline: you pay for your tools once, reach them directly, and let what works become something you — and your team — never have to figure out twice.

Install

This repo is a dual-format marketplace — the same repo installs into both Claude Code and Codex (the SKILL.md is shared; each agent reads its own manifest).

Claude Code

claude plugin marketplace add modiqo/rote-skills
claude plugin install rote-onboard@rote-skills

Or one line: claude plugin marketplace add modiqo/rote-skills && claude plugin install rote-onboard@rote-skills

Then run it:

/rote-setup

Installed at user scope by default (available in every project). Use --scope project to scope it to one repo.

Codex

codex plugin marketplace add modiqo/rote-skills
codex plugin add rote-onboard@rote-skills

Or one line: codex plugin marketplace add modiqo/rote-skills && codex plugin add rote-onboard@rote-skills

Then, inside a Codex session, invoke the skill with:

$rote-setup

Invocation differs by agent: Claude Code uses the slash command /rote-setup; Codex uses the skill syntax $rote-setup (Codex skills are invoked with $, not /).

What the wizard does

A guided, interactive wizard that takes you from zero to a working rote install:

1. Installs rote if missing — CLI one-liner (curl -fsSL https://getrote.dev/install | bash) or a VS Code-family editor extension (VS Code Marketplace / Open VSX for Cursor & Antigravity), detecting which editors you actually have.
2. Signs you in — every experience is identity-gated, so sign-in always runs with Google or GitHub before the setup fork.
3. Forks on how far to go — stop at just the CLI, pull curated powerpack adapters, or build adapters from the 872-API built-in catalog (rote adapter catalog search → rote adapter new). All branches run under your signed-in identity.
4. Menu-driven setup — install adapters à la carte from the live registry (with their flows), wire credentials, connect Google OAuth, install the agent skill, and explore — you pick what runs.
5. Value-proof closer — ends by running one live flow against your own data (you pick it, the wizard reads its parameters and asks you for values), so setup finishes with real output, not just "complete."

Every branch is a clear choice; it never silently runs the whole one-liner.

Decision flow

The flow is identity-gated (sign-in always runs) and choice-driven at every fork — the wizard pauses for an AskUserQuestion at each branch and never runs the whole setup silently. Just the CLI exits clean; Powerpack and API catalog both continue on to credentials, the agent-skill install, and a live-flow finale.

Updating

Claude Code:

claude plugin update rote-onboard@rote-skills

Codex: refresh the marketplace snapshot, then reinstall:

codex plugin marketplace upgrade rote-skills
codex plugin add rote-onboard@rote-skills

Skills in this plugin

The rote-onboard plugin bundles five skills (one install pulls them all). They sit alongside the day-to-day rote skill — rote is for using rote; these are for setting up, sharing, and managing it. They're a sequence, not a grab-bag: each one ends by naming the next logical step, so the path becomes muscle memory. See plugins/rote-onboard/INDEX.md for the full map — order, when to reach for which, and how they chain.

Skill	Claude Code	Codex	What it does
rote-setup	/rote-setup	$rote-setup	First-run setup: install → sign in → fork (stop / powerpack / build an adapter) → prove it works. The front door; delegates adapter work to the skills below.
rote-adapter-create	/rote-adapter-create	$rote-adapter-create	Dry-run-first adapter creation from the catalog, a web-found spec, or a local file. Analyzes the spec, then pre-fills auth + toolsets — and when auth is ambiguous, researches the provider's docs to recommend a scheme with inline setup steps (API key / OAuth2).
rote-adapter-config	/rote-adapter-config	$rote-adapter-config	Tune an existing adapter — auth, base-url, write guard, sensitivity, capability index, re-auth, GraphQL field filter, policies, subagent.
rote-registry	/rote-registry	$rote-registry	Share artifacts to the registry. At the moment an adapter is minted or a flow crystallized, checks whether it already exists in your orgs, pushes at a chosen visibility, then surfaces members and offers invites. Also shows usage/quota across all your orgs.
rote-update	/rote-update	$rote-update	Update the rote binary and the skills together.

Each skill closes a clean run with a dry one-liner keyed to that run's facts — a small, honest reminder that rote talks straight to the provider's API, with no metered middleman proxy in the path. The shared rules for that live in the INDEX above.

License

MIT