The chef's standard. One skill. Craftsman's taste. Zero AI slop.

Trust the chef — state the goal. Omakase applies the bar, remembers taste, and critiques before anything significant ships.

What it is

Omakase is a portable quality standard for agent work: twelve rules, a critique rubric, and project memory (.omakaseagent/taste.md, decisions.md). Not a bag of prompts.

Generic, hedging, over-engineered output fails. Non-trivial work explains why this approach.

Quick start

Paste this into your agent (Claude Code, Cursor, wherever it lives) — the setup is agentic, not a script:

Run `npx omakaseagent init` in this repo. Then read the AGENTS.md and Omakase
skill it installs, and adopt the twelve rules and the critique gate for all
work here from now on. When you're done, tell me what standard you're now
working under and how I should hand you tasks.

Prefer the terminal? npx omakaseagent init — but the agent should be the one reading and adopting what gets installed.

Reload your harness, then talk to a lead (not a menu of skills):

Harness	Example
Cursor / Claude Code	`@omakase-engineer add rate limiting with backoff`
OpenCode	`opencode run --agent omakase-engineer "add rate limiting…"`
Grok	`grok --agent omakase-engineer "…"`
Claude CLI	`claude -p --agent omakase-critic "review the auth module"`

Leads: @omakase-engineer · @omakase-critic · @omakase-archivist — specialists stay internal; leads delegate.

Router skill (omakase-router) is only for plan, taste, handoff, and init when you are not using a native lead:

/omakase-router plan <goal>
/omakase-router critique <target>

Harness quirks, Codex names, and troubleshooting: docs/NATIVE-SUBAGENTS.md.

The standard (always loaded)

File	Role
OMAKASE-RULES.md	Twelve non-negotiable rules
OMAKASE-CRITIQUE.md	Rubric for major output
OMAKASE-PRINCIPLES.md	Why this is a standard, not a prompt pack

Memory lives in .omakaseagent/ after init.

Install

New project:

npx omakaseagent init

Skill + agents only (harness already configured):

npx omakaseagent skills install          # auto-detect harness
npx omakaseagent skills install cursor   # or: claude | agents | grok | codex

Also on the skills ecosystem: npx skills add rblalock/omakaseagent (skills.sh).

Use npx omakaseagent skills install --no-native-agents for skill-only. Verify artifacts: npm run verify:native-agents.

Update an existing install

Omakase ships skill trees and native agent prompts from the npm package. After a new release (see npm), refresh each project that should pick up the change.

Per project (recommended — skill + native agents for all harnesses this repo uses):

npx omakaseagent@latest init

One harness only (e.g. you only use Grok in this repo):

npx omakaseagent@latest skills install grok   # cursor | claude | agents | codex

Global install (user-level skill/agents dir):

npx omakaseagent@latest skills install grok --global

Then start a new harness session (reload Cursor, new grok run, etc.) so updated prompts load.

init is safe to re-run: it refreshes .grok/, .claude/, .agents/, and related bundles from dist/. It does not overwrite existing .omakaseagent/taste.md or decisions.md.

If this repo lists omakaseagent as a dependency, bump or npm update omakaseagent first, then run init or skills install from the project root.

Check the installed CLI version: npx omakaseagent --version (or npx omakaseagent with no args for help).

Developing this repo

Source: skill/. Shipped bundles: dist/ (committed — do not edit by hand).

npm link              # local CLI
npm run build
npm run verify:native-agents
npx omakaseagent init      # dogfood in this clone

License

Apache 2.0

Omakaseagent

README