Deck No. 09 · Project

Agent Product

An MCP server or agent feature you can host, measure and hand to customers.

A product with an agent inside, or a tool for other people's agents: model calls through the Claude API, tools exposed as an MCP server, an HTTP service on Cloudflare Workers, and a skill that teaches customers' agents how to use it.

Who it's forDevelopers who have used agents and now build one into a product, or wrap their API so that other agents can call it. You should be comfortable with TypeScript or Python and have an Anthropic API key and a Cloudflare account. Model calls, evaluation runs and skill comparison runs spend API credits; the cards say where.
npx skills add anthropics/skills@e4bb5d960d87126ff3b3a8af1a104b7db67560ba --skill claude-api mcp-builder skill-creator -g -y && \
npx skills add muratcankoylan/Agent-Skills-for-Context-Engineering@d9202c76d6dc8351e648badb576a6e9b5aae1e7f --skill multi-agent-patterns -g -y && \
npx skills add honojs/skills@e54ce51f6effaeb574ebd2629e89e086e14a2db3 --skill hono -g -y && \
npx skills add cloudflare/skills@5108325ecceabc899c761e07aa617b09c02445ee --skill cloudflare -g -y && \
npx skills add getsentry/skills@a582c51ee29d0f35e6ac321340796f77d43da733 --skill skill-scanner -g -y

Fetches 7 skills from 5 author repos, each pinned to the version we reviewed. Nothing is mirrored. Uses the open-source skills CLI; needs Node.js 22.20 or newer.

What's inside

7 skills. Each card links to its author's repository at the exact version we reviewed.

  1. Claude API 1/7 Official Claude API Writes Claude API and Managed Agents code from bundled references for seven SDK languages and cURL instead of from memory, with guided model migration, cost and eval workflows. @anthropics Apache-2.0

    Claude API

    The model side. It writes Claude API code from bundled references instead of from memory, with current model names, and has guided workflows for cost and evals.

    Official · Anthropic Apache-2.0
    Source anthropics/skills at e4bb5d9
  2. Multi-agent patterns 2/7 Agent architecture Ready Design reference for multi-agent systems: when extra agents are justified, supervisor, swarm and hierarchical layouts, context isolation, consensus and failure handling. @muratcankoylan MIT

    Multi-agent patterns

    Read it before you draw boxes. It asks whether the product needs more than one agent at all, and which layout fits if it does.

    Works right away Community · Muratcan Koylan MIT
    Source muratcankoylan/Agent-Skills-for-Context-Engineering at d9202c7
  3. MCP builder 3/7 Official MCP servers Needs app Guides the agent through building an MCP server in TypeScript or Python: plan the tools, implement them with schemas and clear errors, test, then write evaluations. @anthropics Apache-2.0

    MCP builder

    The tool side. Plan the tools, give them schemas and clear errors, then write evaluations that show whether a model can use them.

    Needs an app or CLI Official · Anthropic Apache-2.0
    Source anthropics/skills at e4bb5d9
  4. Hono 4/7 Official API framework Needs app Inline reference for Hono routing, middleware, validation, typed client and streaming, plus the Hono CLI to list routes and send test requests to the app without starting a server. @honojs MIT

    Hono

    A remote MCP server or an agent endpoint is an HTTP service. Hono runs on Workers without an adapter, and its CLI lets the agent call a route in-process.

    Needs an app or CLI Official · Hono MIT
    Source honojs/skills at e54ce51
  5. Cloudflare platform 5/7 Official Cloud platform Maps what you want to build to the right Cloudflare products, with bundled guides for D1, KV, R2, Queues and Hyperdrive, and checks current docs before it trusts limits or prices. @cloudflare Apache-2.0

    Cloudflare platform

    Where it runs. It picks the Cloudflare products that fit the job and carries the storage guides for whatever state your agent has to keep.

    Official · Cloudflare Apache-2.0
    Source cloudflare/skills at 5108325
  6. Skill creator 6/7 Official Skill authoring Needs app Drafts a skill with you, runs test prompts with and without it, shows the results side by side in a browser viewer and tunes the description so the skill triggers. @anthropics Apache-2.0

    Skill creator

    Your customers' agents need instructions too. This drafts a skill for your product and compares runs with and without it, which gives you evidence to show.

    Needs an app or CLI Official · Anthropic Apache-2.0
    Source anthropics/skills at e4bb5d9
  7. Skill scanner 7/7 Skill security Needs app Checks a skill before you adopt it: a bundled static scan plus a guided read for prompt injection, risky scripts, broad permissions, leaked secrets and supply chain risk. @getsentry Apache-2.0

    Skill scanner

    A skill you publish is text that other people's agents will follow. This reads yours for injection risks, broad permissions and leaked secrets before it leaves your repository.

    Needs an app or CLI Community · Sentry Apache-2.0
    Source getsentry/skills at a582c51

Good to know

What you bring

  • Claude API access: an Anthropic API key or an ant CLI login (usage is billed per token)
  • The Anthropic SDK for your project's language
  • Python 3 or Node.js, only for the optional bundled eval and migration scripts
  • Node.js or Python to build and run the server
  • Optional: Anthropic API key for the bundled evaluation script (paid)
  • Node.js (22.12 or newer for Hono v5)
  • Hono CLI, added to the project as a dev dependency
  • Cloudflare account to create resources and deploy
  • Node.js for Cloudflare's cf CLI or Wrangler
  • Python 3 for the eval, viewer and packaging scripts
  • Claude Code for the description optimiser, which runs claude -p
  • uv to run the bundled Python scanner

Pick one, not both

  • Model layer: this deck calls Claude through Anthropic's SDKs. For a TypeScript product that should stay provider-neutral, take ai-sdk by Vercel in place of claude-api. Both answer requests to write model-calling code, so do not run them together.
  • Local server: an MCP server that only runs on the user's machine needs neither hono nor cloudflare. Leave both out.
  • Hosting: on Vercel, replace the cloudflare card with deploy-to-vercel. Use one host.
  • State: sessions that must survive between requests are a job for durable-objects in the Backend & Cloud pack. Add it when you need it.
  • Skill authoring: writing-skills from Superpowers overlaps with skill-creator. Do not add a second authoring skill.
  • Monitoring: sentry-instrument in the Sentry pack can record the model calls of a deployed product. It keeps prompts and responses in what your app sends to Sentry unless you ask it to limit that.
Review slipDeck No. 09
  • Every skill file read in full
  • Pinned to a reviewed version
  • Author and licence on every card
  • Requirements and costs disclosed
  • No sponsored skills inside
Last reviewed
Nov 2026
Reviewed