Ultracite & Fast Validation Recipe

Zero-config linting preset + a Bun/tsgo/oxlint stack that validates an entire repo in ~250-500ms. Fast enough to run on every line of code changed.

Ultracite

Ultracite is a zero-config preset for ESLint, Biome, and Oxlint. It provides hundreds of production-grade rules for JavaScript/TypeScript projects with framework-specific presets. Source: https://ultracite.ai; https://docs.ultracite.ai

Use Ultracite for all linting/formatting configuration instead of hand-rolling ESLint, Prettier, Biome, or oxlint configs. It handles the opinionated baseline; custom ESLint rules layer on top for repo-specific policy only (see Lint-Enforced Agent Guardrails).

Docs: https://docs.ultracite.ai | GitHub: https://github.com/haydenbleasel/ultracite

Current source snapshot: ultracite@7.8.4 is the npm latest release, published 2026-07-02; the public Git repository default branch was 9e7f4389caaad9b5895b84a7c026e6dbe702155a on review, with no public v7.8.4 tag found. Current docs still position Ultracite as a zero-config preset across Biome, ESLint/Prettier/Stylelint, and Oxlint/Oxfmt, with npx ultracite init flags for package manager, framework, editor, agent, hooks, --install-skill, and type-aware rules. The hook docs preserve the agent-loop invariant: after AI edits, run npx ultracite fix and preserve existing hook config without duplicate entries. Source: npm registry, GitHub, and docs.ultracite.ai, 2026-07-03

Setup

bun x ultracite@latest init \
  --linter oxlint \
  --pm bun \
  --frameworks react next \
  --editors cursor \
  --agents universal claude \
  --hooks cursor claude \
  --type-aware

The CLI detects your lockfile and generates config files, editor settings, agent rule files, and hook configs in one pass. For existing projects, run bun x ultracite@latest init interactively.

Toolchain Selection

Prefer Oxlint for new projects. It handles React, TypeScript, Next.js, Vue, Jest, Vitest, JSDoc, and more out of the box.

Oxlint Configuration

// .oxlintrc.json
{
  "$schema": "./node_modules/oxlint/configuration_schema.json",
  "extends": [
    "./node_modules/ultracite/config/oxlint/core/.oxlintrc.json",
    "./node_modules/ultracite/config/oxlint/react/.oxlintrc.json",
    "./node_modules/ultracite/config/oxlint/next/.oxlintrc.json"
  ]
}

// .oxfmtrc.jsonc
{
  "tabWidth": 2,
  "useTabs": false,
  "semi": true,
  "singleQuote": false,
  "trailingComma": "es5",
  "bracketSpacing": true,
  "arrowParens": "always",
  "printWidth": 80
}

CLI Commands

Agent Integration

Ultracite generates agent-specific rule files so AI assistants inherit lint standards:

• Agent rules (--agents): Generates AGENTS.md, .claude/CLAUDE.md, .cursor/rules/ultracite.mdc - code quality guidance for AI code generation.
• Agent hooks (--hooks): Generates .cursor/hooks.json, .claude/settings.json - runs ultracite fix after every AI edit, formatting and fixing before the agent reviews its own output.
• MCP server: For cloud agents without local access, add to MCP config:

{
  "mcpServers": {
    "ultracite": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://docs.ultracite.ai/mcp"]
    }
  }
}

Monorepo Setup

Config at repo root. One ultracite check covers all packages. With Turborepo, add "//#check": {} and "//#fix": { "cache": false } as root tasks.

Git Hooks

npx ultracite init --integrations husky,lint-staged

Supported: Husky, Lefthook, lint-staged, pre-commit (Python framework).

Reusable Skill

Install once, carry across repos:

npx ultracite init --install-skill
# or
npx skills add haydenbleasel/ultracite

The Fast Validation Recipe

A stack optimized for sub-500ms full-repo validation:

Target Performance

Full repo validation in ~250-500ms. At this speed, validation can run on every line of code changed - not just on commit or CI. This turns linting from a batch check into a real-time feedback signal for both humans and agents.

Rusty's earlier recipe bookmark states the same stack in operational terms: Bun for startup, TS7/tsgo for type checking, opinionated oxlint/oxfmt config, focused unit tests, and agentic hooks. The useful update is the per-line feedback loop, not a new package: when the validation loop is this fast, agents can run checks against each changed line or small edit batch instead of waiting until a PR-level pass. Source: X/@rustydotwtf, 2026-03-10

Agentic Hooks

When validation is fast enough to run in-loop, agents get immediate feedback on every edit. A slow lint pass (5-10s) means the agent batches code and checks later. A fast lint pass (250ms) means the agent catches mistakes on the line it just wrote.

This compounds with Lint-Enforced Agent Guardrails. Custom rules that detect slop patterns + sub-second validation = agents that cannot produce slop without immediately knowing about it.

Why Not Hand-Rolled Configs

Ultracite handles the hundreds of standard rules. Custom ESLint rules handle what's unique to your codebase (mock-echo detection, query-key-deps, architecture policy). This is the correct split.

---

Timeline

• 2026-07-04 | Reviewed Rusty's text-only fast-validation recipe row as supporting evidence for the existing Bun + tsgo + oxlint/oxfmt + focused-test + agent-hook stack. No new tool or version snapshot was needed because the durable source authority remains Ultracite/npm/GitHub; the row sharpens the "run checks on every changed line" operator model. Source: X/@rustydotwtf, 2026-03-10
• 2026-07-03 | Deep-reviewed the Rusty bookmark as Ultracite source evidence and refreshed package/source authority: ultracite@7.8.4 on npm, repo HEAD 9e7f4389caaad9b5895b84a7c026e6dbe702155a, no matching public v7.8.4 Git tag found. Source: X/@rustydotwtf, 2026-03-10; Source: npm/GitHub/docs, 2026-07-03
• 2026-06-17 | Added inline provenance and timeline so the recipe can be audited separately from the executable tooling choices. Source: https://ultracite.ai; https://docs.ultracite.ai
• 2026-04-07 | Created fast-validation recipe from Kevin's manual notes, Ultracite docs, and the JS toolchain direction. Source: manual (Kevin Liu, X post); https://ultracite.ai