diff --git a/README.md b/README.md
index 0dabd63..226221b 100644
--- a/README.md
+++ b/README.md
@@ -100,7 +100,7 @@ The `plan` and `fetch-sources` steps are returned from checkpoints. Only `write-
 Make every AI session in your project use Tidebase correctly:
 
 ```bash
-npx tidebase init        # writes a Tidebase section into AGENTS.md/CLAUDE.md (idempotent)
+npx @tidebase/sdk init        # writes a Tidebase section into AGENTS.md/CLAUDE.md (idempotent)
 ```
 
 Give your assistant direct access to runs, gates, and recovery via MCP:
diff --git a/packages/sdk/src/cli.ts b/packages/sdk/src/cli.ts
index a1a228a..9a1ab99 100644
--- a/packages/sdk/src/cli.ts
+++ b/packages/sdk/src/cli.ts
@@ -79,6 +79,6 @@ const command = process.argv[2]
 if (command === 'init') {
   init()
 } else {
-  console.log('Usage: tidebase init')
+  console.log('Usage: npx @tidebase/sdk init')
   if (command) process.exitCode = 1
 }
diff --git a/packages/tidebase-cli/.npmignore b/packages/tidebase-cli/.npmignore
deleted file mode 100644
index e69de29..0000000
diff --git a/packages/tidebase-cli/cli.js b/packages/tidebase-cli/cli.js
deleted file mode 100755
index f5c0c17..0000000
--- a/packages/tidebase-cli/cli.js
+++ /dev/null
@@ -1,79 +0,0 @@
-#!/usr/bin/env node
-/**
- * tidebase CLI — currently one command:
- *
- *   tidebase init   Write a Tidebase usage block into the project's
- *                   AGENTS.md/CLAUDE.md so AI coding sessions in this repo
- *                   use Tidebase correctly instead of hand-rolling status
- *                   columns. Idempotent: skips files that already have a
- *                   "## Tidebase" section.
- */
-import { readFileSync, writeFileSync, existsSync } from 'node:fs';
-import { resolve } from 'node:path';
-const SNIPPET = `## Tidebase (checkpoint layer)
-
-This project uses [Tidebase](https://github.com/BlueprintLabIO/tidebase) (\`@tidebase/sdk\`) to checkpoint long-running agent workflows. Server: \`$TIDEBASE_URL\` (default \`http://localhost:7373\`), Studio dashboard on \`:5173\`.
-
-**When writing multi-step or long-running workflows in this repo, wrap them in Tidebase instead of hand-rolling status columns, retry flags, or checkpoint blobs:**
-
-\`\`\`typescript
-import { Tidebase } from '@tidebase/sdk'
-const tide = new Tidebase()
-
-await tide.run('workflow-name', { runId }, async (run, input) => {
-  const a = await run.step('step-a', () => doA(input))      // checkpointed
-  await run.state.patch({ status: 'working', progress: 0.5 }) // live UI state
-  return run.step('step-b', () => doB(a))
-})
-\`\`\`
-
-Rules:
-
-- One \`run.step('name', fn)\` per unit of work you must not repeat (LLM calls, external writes). Re-invoking a run with the same \`runId\` replays completed steps from checkpoints — never re-executes them.
-- Steps with external side effects MUST declare a resume contract: \`run.step('send', { sideEffects: ['email.send'], idempotencyKey: \\\`send:\${id}\\\` }, fn)\`. Without an idempotency key, failures park in \`manual_review\` instead of auto-retrying.
-- Human approval before a risky action: \`const d = await run.gate('approve-x', { prompt })\`; check \`d.decision === 'approved'\`. Gates are durable and resolve exactly once.
-- Parallel subagents: \`run.fanout(name, branches)\` — child runs are idempotent by branch name on resume.
-- Record token usage with \`run.usage.record({ kind: 'llm', model, inputTokens, outputTokens, costUsd })\` after LLM calls.
-- Do NOT build separate run-status tables, progress columns, or retry-flag columns — query Tidebase (\`GET /runs\`, \`GET /runs/:id\`) instead.
-- Debugging a stuck/failed run: check Studio or \`GET $TIDEBASE_URL/runs/:runId\` for steps, failure classification, and pending gates.
-`;
-function inject(file) {
-    const path = resolve(process.cwd(), file);
-    if (!existsSync(path))
-        return 'created';
-    const current = readFileSync(path, 'utf8');
-    if (/^## Tidebase\b/m.test(current))
-        return 'skipped';
-    writeFileSync(path, `${current.replace(/\n*$/, '\n\n')}${SNIPPET}`);
-    return 'written';
-}
-function init() {
-    const targets = ['AGENTS.md', 'CLAUDE.md'].filter((f) => existsSync(resolve(process.cwd(), f)));
-    if (targets.length === 0) {
-        writeFileSync(resolve(process.cwd(), 'AGENTS.md'), SNIPPET);
-        console.log('created AGENTS.md with the Tidebase section');
-    }
-    else {
-        for (const file of targets) {
-            const result = inject(file);
-            console.log(result === 'skipped'
-                ? `${file}: already has a "## Tidebase" section, skipped`
-                : `${file}: appended the Tidebase section`);
-        }
-    }
-    console.log(`
-Next steps:
-  - Server + Studio: git clone https://github.com/BlueprintLabIO/tidebase && cd tidebase && docker compose up -d postgres && pnpm install && pnpm dev
-  - Give your AI assistant run access (MCP):
-      claude mcp add tidebase -e TIDEBASE_URL=http://localhost:7373 -- npx -y @tidebase/mcp
-  - Docs for agents and humans: https://tidebase.dev/llms.txt`);
-}
-const command = process.argv[2];
-if (command === 'init') {
-    init();
-}
-else {
-    console.log('Usage: tidebase init');
-    if (command)
-        process.exitCode = 1;
-}
diff --git a/packages/tidebase-cli/package.json b/packages/tidebase-cli/package.json
deleted file mode 100644
index 09ea20b..0000000
--- a/packages/tidebase-cli/package.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "name": "tidebase",
-  "version": "0.5.0",
-  "description": "CLI shim for Tidebase — `npx tidebase init` writes a Tidebase usage section into AGENTS.md/CLAUDE.md so AI coding agents use the checkpoint layer correctly. Full SDK: @tidebase/sdk",
-  "license": "Apache-2.0",
-  "type": "module",
-  "bin": {
-    "tidebase": "cli.js"
-  },
-  "files": [
-    "cli.js"
-  ],
-  "scripts": {
-    "prepublishOnly": "cp ../sdk/dist/cli.js ./cli.js"
-  },
-  "homepage": "https://tidebase.dev",
-  "repository": {
-    "type": "git",
-    "url": "https://github.com/BlueprintLabIO/tidebase"
-  },
-  "keywords": ["tidebase", "ai-agents", "checkpoint", "cli"]
-}