One command plants a complete bun workspace: Next.js or SvelteKit, Expo, Elysia or Hono, Convex — official generators, stitched together, nothing vendored.
x18n reads the locale files of any project, finds where each key is used, and reports missing, unused and broken keys. A CLI, a language server, a VS Code extension, a UI and MCP tools share one engin
TraceRoot is an open-source, agent-native improvement layer for AI agents. Bring production traces, detector signals, and evals into your coding workflow.
Taiwan-built, self-hosted AI agent harness in a single Go binary — writes its own tools, sandbox-tests them, and shares them over MCP with Claude Code and Codex.
🌱 Plant a bun-first Turborepo and grow it — web, mobile, API and Convex backend from one command.
翻譯準備中
取自儲存庫 README
README
正在載入完整 README。 以下為已儲存的摘錄。
Plant a bun-first Turborepo. Grow the apps you actually want.
🌱 v2 is in prerelease (bunx create-groot@next). It adds adoption of existing projects, previewable capability plans, journaled apply/resume/rollback, evidence-based verification, managed agent instructions, tasks for installed coding agents, and an MCP server — see what's new.
Plant a bun-first Turborepo. Grow the apps you actually want.
🌱 v2 is in prerelease (bunx create-groot@next). It adds adoption of existing projects, previewable capability plans, journaled apply/resume/rollback, evidence-based verification, managed agent instructions, tasks for installed coding agents, and an MCP server — see what's new.
🏛️ v1 is stable (bunx create-groot@latest). The stability contract is binding: the command set, every documented flag, exit codes, and the groot.json schema follow semver. v2 keeps every v1 command, flag, and exit code; its deliberate changes are listed in stability.md.
What is groot?
groot is a Bun-first CLI for humans and coding agents that creates, adopts, evolves, and verifies products.
It started as a scaffolder, and that is still its foundation: it plants a Turborepo monorepo (or a single app) and grows it with the apps you pick — a web app, a mobile app, a desktop app, an API, and a backend — in one command. Instead of shipping its own frozen templates, groot orchestrates each framework's official generator (create-next-app, sv create, create-expo-app, create-tauri-app, create-hono, …) and then stitches the results into one coherent bun workspace: shared config packages, workspace protocol dependencies, non-conflicting dev ports, a clean root turbo.json, and a single lockfile.
Plant a new workspace — interactive, or fully flag-driven (--dry-run --json for agents); --topology single plants one app at the project root (v2)
groot add <scaffold>
Grow an existing workspace: groot add hono, groot add sveltekit --path apps/marketing (v2 workspaces get a free dev port automatically)
groot init my-app --github
Plant, then create + push a private GitHub repo via gh (--public for public); absent/unauthenticated gh degrades to printed steps
groot doctor
Health checks with suggested fixes — ports, lockfiles, manifests, per-framework invariants
The full contract — every flag, exit code, and the groot.json manifest schema — lives in docs/cli-spec.md; docs/stability.md defines what's covered by semver (binding since v1.0). Running groot in CI or from an agent? Recipes in docs/ci.md.
v2: create, adopt, evolve, verify
v2 (prerelease) turns every change into a plan you can read before anything is written, executes it with a journal you can resume or roll back, and proves the result with evidence tied to the exact revision it checked.
bunx create-groot@next init my-api --topology single --api hono --yes
cd my-api
groot plan add auth # data (Drizzle + SQLite) is added first — auth requires it
groot apply <planId> # journaled; refuses if a file it touches changed since planning
groot verify --profile all # structural · build · runtime · product-flow (sign-up → protected data → 401)
groot context sync # managed AGENTS.md / CLAUDE.md sections + the groot skill
Already have a project? groot inspect reads it without executing anything, and groot adopt registers it — writing only groot.json and groot.lock.json, never moving your files or touching dirty or staged work.
groot plan add <capability> · groot plan context-sync
Preview every file write/edit, dependency, command, environment variable, verification, and recovery limit — nothing changes yet
groot apply <plan> · groot status · groot resume <op> · groot rollback <op>
Execute with stale-plan detection, one writer at a time, crash-safe resume, and rollback that refuses to overwrite your later edits
groot verify · groot evidence
Run verification profiles and inspect the recorded, redacted evidence
groot context [--task "…"] · groot context sync
Task-scoped project knowledge (variable names, never values) · managed agent instructions that preserve your own text
groot task … · groot review <id>
Hand bounded work to an installed coding agent (Claude Code; Codex adapter included) in an isolated worktree, review it, integrate it with fresh checks
groot mcp
The same operations as typed MCP tools over stdio, for any MCP client
groot schema [name]
Machine-readable contracts for every --json document
Certified capabilities: data (data.drizzle-sqlite) and auth (auth.better-auth, email + password with protected example data) on Hono + Bun, in single-app and monorepo projects, created or adopted. Everything with --json returns one result envelope with stable GROOT_E_* error ids. The normative v2 contract is docs/v2-cli-spec.md; the design is docs/v2-architecture.md; what is shipped versus planned is tracked in the v2 ledger.
How it works
Generate — groot runs each framework's official generator with pinned, non-interactive flags. No vendored templates, so scaffolds never rot behind upstream.
Stitch — groot patches the output into monorepo shape: workspace globs, @repo/* package names, shared TypeScript config, deterministic dev ports, merged .gitignore, one root lockfile.
Verify — structural checks and the root install prove the workspace coheres; groot doctor re-runs health checks on any groot workspace, any time, with suggested fixes.
create-turbo plants an excellent tree — two Next.js apps and shared packages. groot is for when your monorepo needs to look like your product on day one: Next.js and Expo and an Elysia API and a Convex backend, already wired together, with the boilerplate decisions (ports, package names, shared configs) made consistently. groot uses create-turbo under the hood for the trunk, then grows the branches.
Project status
Phase 0 — production-grade OSS repository: CI, CodeQL, OpenSSF Scorecard, release automation, signed provenance, this documentation.
v0.1 — create-groot published to npm (OIDC provenance), 5-platform binaries + checksum-verified installers live.
v0.2 — groot init with the full scaffold matrix: the resolve → preflight → generate → stitch → verify pipeline, interactive + fully non-interactive, --dry-run --json, real-generator E2E in CI.
v0.3 — groot add (grow an existing workspace, with occupancy rules and targeted rollback) and groot doctor (health checks with fixes).
v1.0 — the stability contract goes binding: CLI surface, exit codes, and manifest schema under semver — plus SBOM + build-provenance attestations on every release and the live docs site.
v2.0 (prerelease next) — adopt existing projects, previewable plans with journaled apply/resume/rollback, certified auth + data capabilities with product-flow evidence, managed agent instructions, installed-agent tasks with review and integration, and an MCP server. Progress and remaining gates: v2 ledger.
PRs are very welcome — especially scaffold adapters for new frameworks. Start with CONTRIBUTING.md. Every PR is reviewed by CI and by Greptile; merges require a 5/5 review score.
