Skip to main content
skill CX Codex

embedded-agents-agent-sdk — Claude Code skill: setup and guide

embedded-agents-agent-sdk — a skill for Claude Code and Codex to embed Claude agents into a Next.js app using Claude Agent SDK. Runs when Agent SDK integration is requested or when an agent requires Claude Code engine tools.

MIT ★ 741 updated 2026-11-14

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add phuryn/product-engineering
/plugin install app-stack@product-engineering
Open on GitHub

What embedded-agents-agent-sdk does

This skill integrates the Claude Code runtime into Next.js applications via @anthropic-ai/claude-agent-sdk. The SDK launches the engine as a subprocess, giving the agent built-in shell access, file operations, and the Claude Code system prompt. Each agent lives in .claude/agents/<name>/ with instructions isolated strictly inside AGENTS.md. A generic askAgent wrapper handles execution without hardcoded identifiers.

The setup supplies template files for src/lib/agent.ts, a streaming route at src/app/api/agents/route.ts, and Vitest test suites. MCP servers and local skills map directly within each agent directory, reading secret variables from the host environment rather than raw files.

A persistent Node host environment is mandatory. Running the wrapper requires ANTHROPIC_API_KEY, alongside zod, server-only, and Vitest. On a local development machine without an API key, the SDK automatically falls back to an active Claude CLI login. Example research agents require an additional PERPLEXITY_API_KEY.

What it helps with

  • Embed Claude Code runtime subprocesses inside Next.js App Router applications
  • Organize multi-agent instructions and MCP configurations within isolated directories
  • Stream agent responses via Server-Sent Events to custom chat frontends
  • Validate agent contracts with contract unit tests and live model Vitest suites

Example prompts for the agent

  • Embed the Claude Agent SDK wrapper into this Next.js app
  • Add a new embedded agent with local knowledge files using embedded-agents-agent-sdk
  • Configure a streaming SSE route for Claude Code agents

How to install embedded-agents-agent-sdk

1Plugin marketplace

/plugin marketplace add phuryn/product-engineering
/plugin install app-stack@product-engineering

This installs the whole app-stack plugin with all its skills

2Skills CLI (Vercel)

npx skills add phuryn/product-engineering --skill embedded-agents-agent-sdk -a claude-code

3Manually: put the files in place

~/.claude/skills/embedded-agents-agent-sdk/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/embedded-agents-agent-sdk/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about embedded-agents-agent-sdk

How to install embedded-agents-agent-sdk in Claude Code?

Run npx skills add phuryn/product-engineering --skill embedded-agents-agent-sdk -a claude-code or install the marketplace plugin via /plugin marketplace add phuryn/product-engineering followed by /plugin install app-stack@product-engineering.

What does embedded-agents-agent-sdk do?

It provisions the askAgent wrapper, Next.js streaming API endpoints, and directory scaffolds to run Claude Code agents inside a web service.

What is required to run embedded-agents-agent-sdk?

It needs a persistent Node runtime, ANTHROPIC_API_KEY, Next.js, and npm packages @anthropic-ai/claude-agent-sdk, zod, and server-only.

Source text

SKILL.md · MIT
---
name: embedded-agents-agent-sdk
description: "Only when the Agent SDK is asked for, an agent needs Claude Code's own tools, or the app already runs it: embed Claude agents in a Next.js app with the Agent SDK."
---

# Embedded agents — Claude Agent SDK

**Use the Agent SDK when the agent needs the Claude Code engine.** Also use it when the product already runs it. The SDK starts that engine as a subprocess with built-in file and shell tools and Claude Code's own system prompt. It loads skills and MCP and needs a long-lived Node host. Both embedded-agent skills run the same agent directories. The `embedded-agents-messages-api` skill is the default for a new product. It uses the Anthropic SDK's tool runner in the app's own process on any Node host, with only the wrapper's tools and no subprocess.

**Read the product's own agent instructions before changing its embedded agents.** Start with its root `AGENTS.md` or `CLAUDE.md`, and tell any subagent spawned for this work to read it too. Its rules on secrets, environment variables, validation, callers, tests and release apply here. This skill adds how to run Claude agents inside the app. Read `reference/playbook.md` before changing the wrapper's contract or explaining it in depth. It holds the full playbook and the measurement behind every option.

**Keep request, caller, storage and UI ownership in the app.** The agent owns reasoning and tools. An agent is a directory, `.claude/agents/<name>/`, with `AGENTS.md` as its only instruction file. It may also hold `knowledge/`, agent-local skills in `.claude/skills/`, and `.mcp.json`. One wrapper function, `askAgent`, runs any agent. No agent name appears in code.

## 1. First use: install the wrapper

**Search for `askAgent` before installing another wrapper.** If it exists, skip to § 2. Otherwise:

1. **Create only the smallest requested app when none exists.** Use Next.js App Router in TypeScript, which these files target. Add nothing unasked for: no database, sign-in or billing for a local prototype.
2. **Install the missing dependencies.** Run `npm i @anthropic-ai/claude-agent-sdk zod server-only` and `npm i -D vitest`.
3. **Copy the wrapper, route and tests, changing only import paths.**

   | From this skill | To the app |
   |---|---|
   | `templates/agent.ts` | `src/lib/agent.ts` — the wrapper |
   | `templates/route.ts` | `src/app/api/agents/route.ts` — `GET` lists agents, `POST` streams one |
   | `templates/agent.test.ts` | `tests/unit/agent.test.ts` — the contract, inside `npm test`, no model call |
   | `templates/agents.live.test.ts` | `tests/live/agents.live.test.ts` — every agent against the real model |

   The tests import `@/lib/agent`, so Vitest needs the same `@` alias as `tsconfig.json`. The live test gets its own Vitest config (include `tests/live/**/*.live.test.ts`, `testTimeout: 180_000`, `fileParallelism: false`) and a script, `agents:live`. It costs money, so it stays out of `npm test` and the gate.
4. **Set each required secret per environment and never commit it.** Set `ANTHROPIC_API_KEY` and every secret referenced by an agent's `.mcp.json`; apply `environments`. On a developer's laptop the SDK falls back to their own Claude login when the key is absent; a deployed product always uses the key.
5. **Keep the contract.** Every option in `agentOptions()` exists because an SDK default failed a measurement (playbook § 9). Change one only after re-measuring it.

## 2. Add an agent

1. **Create the agent directory and fill in its instructions.** Run `node <skill dir>/scripts/new-agent.mjs <name> [--knowledge] [--mcp]`, then fill in `AGENTS.md`: role, sources, output format, refusals. A fixed sign-off line is the cheapest live-test proof that the instructions are in force.
2. **Keep knowledge and automatically discovered skills inside the agent directory.** Put knowledge in `knowledge/*.md`, and tell the agent in `AGENTS.md` to answer only from them. Agent-local skills are found automatically in `.claude/agents/<name>/.claude/skills/<skill>/SKILL.md`.
3. **Allow only MCP tools the agent's job needs.** Put servers in `.mcp.json`, with secrets as `${VAR}`, never literal. Run `node --env-file=.env <skill dir>/scripts/discover-mcp-tools.mjs <name>` to list each server's tools, then restrict its `allowedTools`. Anything unlisted is denied. Commit the agent's `.mcp.json`, which holds no secrets. If `git check-ignore` reports it ignored, the repository's `.gitignore` has an unanchored `.mcp.json` rule meant for the root file. Anchor it as `/.mcp.json`.
4. **Discover agents from their directories rather than hardcoding the UI list.** `listAgents()` finds the directory, `GET /api/agents` returns it, and the UI's picker shows it when users choose between agents. Fix a UI that hardcodes that list.
5. **The instruction file is always `AGENTS.md`.** Asked for another name, create `AGENTS.md` and say why: the wrapper loads only `AGENTS.md`, and the SDK loads some other instruction files on its own, which would give the agent two sets of instructions.
6. **Add a live test for each promise worth pinning.** Examples include citing, refusing and answering only from its files. Put it beside the generic test.

## 3. Examples — only when asked

**Copy finished examples only when asked for or described.** The `examples/` agents are: answering only from local product-management knowledge files (`knowledge-agent`), or live-web research through Perplexity with citations (`research-agent`). Copy `examples/<name>/` to `.claude/agents/<name>/` and `examples/<name>.live.test.ts` to `tests/live/`. `research-agent` needs `PERPLEXITY_API_KEY`. Otherwise, never copy them.

## 4. The chat UI

**Build the chat UI for this product and its agent flow.** There is no template. It must do the following:

- **Fill the picker from `GET /api/agents` when users choose agents.** A product may run one front agent and show no picker; the route still checks the name against `listAgents()`.
- **Send the conversation to the route and read server-sent events.** `POST { agent, prompt, history }` to `/api/agents` and read: `{ text }` chunks, then `{ done, agent, metadata }` or `{ error, agent }`. The parser is in playbook § 5.
- **Store the agent on every message; identify replies with several agents.** When there is more than one agent, show it on each reply, so the reader always knows which agent answered. History is the conversation so far, kept in client state unless the product asks to persist it.
- **Render agent output as plain text or sanitized markdown.** Never render raw HTML: it is untrusted text, like anything a user typed. Show an `error` event instead of spinning forever.

## 5. Explaining routing and the tool boundary

**Validate the requested agent before routing it to the wrapper.**

1. UI
2. `POST /api/agents { agent }`
3. the route validates the body and requires the name to be in `listAgents()`
4. `askAgent`
5. `agentOptions()` accepts only a plain directory name and resolves `.claude/agents/<agent>/`
6. `query()` starts the Claude Code engine with that directory as `cwd` and its `AGENTS.md` appended to the system prompt
7. events stream back.

**Keep every measured layer of the tool boundary in force.** See playbook § 9:
1. **Expose only built-in tools on the allowlist.** In `tools`, only those built-ins exist — `Read`, `Glob`, `Grep` by default. `Bash`, `Write`, `Edit` and `WebFetch` do not.
2. **Allow an MCP tool only when its server lists it.** With `permissionMode: 'dontAsk'` and `allowedTools`: an MCP tool runs only if its server's `allowedTools` in `.mcp.json` lists it.
3. **Load only the agent's own MCP servers.** `strictMcpConfig`: only the agent's own `.mcp.json` servers load — nothing from the developer's or the server's user configuration.
4. **List only the agent's own skills.** `skills` enforces this.
5. **Block paths outside the agent directory and every `.env` file.** The PreToolUse guard enforces this.
6. **Cap the run by turns and budget.** Use `maxTurns` (10 by default) and `maxBudgetUsd`.

## 6. Before it leaves localhost

- **Choose a long-lived Node host before the agent leaves localhost.** The SDK runs the Claude Code engine as a subprocess, so an agent cannot run in a serverless function such as a Netlify Function (playbook § 7). A product whose site deploys there needs a second, long-lived Node host (a container or VM, such as Railway), chosen before the agent goes beyond localhost; write its deploy procedure into the product's docs the first time it is used. `next dev` needs nothing.
- **History is a client claim.** `templates/route.ts` accepts `history` from the client, `assistant` turns included, and `composePrompt` turns it into a "Previous conversation" preamble — so a client can forge earlier turns, the agent's own included. That is acceptable for a read-only agent, which can only be talked into saying something to the person forging it. Before any tool acts on the caller's behalf, keep the transcript on the server and stop reading it from the request.
- **Resolve the caller and cap cost before opening the route.** Before anyone but the developer can reach the route, resolve the caller there (who they are, which workspace, which permission: `saas-authorization`) and set a per-actor limit and a `maxBudgetUsd`.
- **Run live tests before releasing any agent, prompt or model change.** Run `npm run agents:live` before any release that adds or changes an agent, its prompt or its model, record the run in the product's test-coverage map (which test proves what, and when it last passed), and describe each agent in its automation doc: what it may read, the exact tools it may call, and what the app enforces around it (`shipping-artifacts`).

## How you can tell it worked

The agent runs on the deployed host, not only on localhost, uses only the tools it was given, and the live agent suite passes against the real model.

## Running the scripts

**Run these scripts from the app's repository root.** `<skill dir>` is this skill's installation location: a plugin, `~/.claude/skills/`, or the repository's own `.claude/skills/`. The scripts write into the current directory.

© Paweł Huryn · MIT · phuryn/product-engineering

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set51

The whole set →
performance-audit-static The performance-audit-static Claude Code skill statically reviews code for performance issues such as N+1 queries, excess data fetching, index gaps, and caching. It runs when a performance audit is requested. CX Codex +1 ★ 741repo stars legal-pages This Claude Code skill drafts legal pages from an app’s behavior and checks them against the code. It comes into play as data handling, cookies, or providers change. CX Codex +1 ★ 741repo stars flight-deck flight-deck — a Claude Code skill that records project state and progress for long-running work. Ask for a status report or where a session left off to trigger it. CX Codex +1 ★ 741repo stars environments The environments Claude Code skill separates development from production and lays out rules for secrets and schema changes. The agent can draw on it when configuring an environment, adding a variable, or handling a leaked key. CX Codex +1 ★ 741repo stars mcp-server This Claude Code skill guides the build, OAuth setup, and directory submission of a remote MCP server. It covers tool tests and checks with real clients. CX Codex +1 ★ 741repo stars transactional-email transactional-email is a Claude Code skill for adding transactional emails and alerts to an app. It covers provider choice, duplicate sends, and one-click unsubscribe. CX Codex +1 ★ 741repo stars grok-agents The grok-agents Claude Code skill runs Grok CLI from another agent as a peer, builder, or reviewer. It comes into play when a task calls for a second model to discuss a decision, take on a development stage, or inspect changes. CX Codex +1 ★ 741repo stars shipping-artifacts shipping-artifacts — a Claude Code skill that builds documentation sets for auditing AI-generated code. It records system architecture, runtime flows, access controls, environment variables, and test coverage. CX Codex +1 ★ 741repo stars

Project sponsors

Companies that keep this analytics open