Make any codebase AI-ready in one command.
Meridian reviews your project and generates the complete AI kit — rules, subagents, skills, slash commands, harness config and real docs — for Claude Code, Cursor, Copilot, Codex and Gemini. Then it keeps the kit fresh as your code evolves.
macOS · Windows · Linux — Node.js ≥ 18 · MIT licensed
$ meridian generate ✔ Reviewed 214 files — read src/cli.ts, src/server/router.ts, +12 more → generating with Claude Code [sonnet] ✔ AI read the codebase → .meridian/docs/codebase-review.md ✔ [1/7] Rules: 1 file (rules.md) ✔ [2/7] Subagents: 4 files (code-reviewer.md, test-fixer.md, …) ✔ [3/7] Skills: 6 files (commit, new-feature, fix-bug, …) ✔ [4/7] Slash commands: 5 files (verify.md, review.md, …) ✔ [5/7] Harness config: 1 file (settings.json) ✔ [6/7] Prompts: 4 files · ✔ [7/7] Docs: 6 files Your project is AI-ready: rules, agents, skills, commands and docs are in place. $ meridian sync --check △ new script: e2e (playwright test) △ new top-level directory: workers Kit is stale — run meridian sync to refresh.
One command writes what you'd never write by hand
meridian generate is AI-first: it digests your codebase — layout,
dependencies, scripts, conventions, source excerpts, a full symbol map — has your best
provider write a real codebase review, and grounds every generated file in it.
Canonical rules, every tool
One rules.md mirrored into CLAUDE.md,
AGENTS.md, GEMINI.md and friends — written from your
real conventions, not a template.
Subagents, skills & commands
Project-tailored reviewers, test-fixers and workflow skills under
.claude/ — including an interactive, approval-gated commit skill with
a hard stop on secrets.
Harness config included
.claude/settings.json pre-approves your project's own test, lint and
build commands and denies reads of .env and key files — fewer
prompts, safer defaults, for the whole team.
No API key required
Signed in to Claude Code, Codex CLI or Gemini CLI? Meridian uses your subscription directly. Or bring an API key, or run local models via Ollama.
A real docs suite
Architecture, conventions, engineer workflow, security, tech debt — the documents a staff engineer would hand a new teammate, derived from your actual code.
Safe by construction
Every AI-suggested path is validated against a per-kind allowlist; failed generations write nothing and resume later; keys live in your OS secret store.
The kit stays alive
Generated config rots the day your code moves on. Meridian records a manifest of what it knew and what it wrote — so the kit can tell when it's stale.
meridian generate
Writes the kit and a manifest: a fingerprint of your project plus a hash of every generated file.
meridian sync
Detects drift — new scripts, frameworks, modules — and refreshes only stale files. Anything you hand-edited is detected by hash and always preserved.
meridian sync --check in CI
Exits 1 when the kit is stale. No AI, no keys needed.
meridian generate ci writes the GitHub Action for you.
Command reference
Run bare meridian for the interactive launcher.
| Command | What it does |
|---|---|
| meridian generate [kinds…] | Review the codebase and generate the complete AI kit |
| meridian sync [--check] | Detect drift and refresh the kit; --check is the CI gate |
| meridian doctor | Detect installed AI tools and missing dependencies |
| meridian install <tool> | Install & configure Claude Code, Cursor, Codex, Gemini CLI… |
| meridian auth [provider] | Store API keys in the OS-native secret store |
| meridian mcp search/install | MCP marketplace — one install configures every detected tool |
| meridian ask "…" | Ask AI via a cost/speed/quality-aware router |