Hands-on portfolio project · Week 4 · Peer-reviewed · module M11 (Extension Triangle). Part of the ForgeMind AI — AI Productivity Essentials course.
Done when: an agent that orchestrates your own Skill (HO6) + your own MCP server (HO8) into one real, unattended workflow.
This is the Week-4 capstone, assembled at week's end after you've built your own MCP server (HO8) and your own agent (HO9). You take the pieces you already built and wire them together:
- an orchestrating agent that runs the whole job from one trigger,
- your HO6 Skill — the repeatable step, in your voice/house style,
- your HO8 MCP server — the fresh data the agent pulls in,
- optionally a Connector to deliver the result (Slack, Gmail, a CRM, etc.).
One trigger then runs your MCP → your Skill → deliver, unattended.
No API key needed. Everything runs on your normal Claude.ai login (Pro or Team), plus the free Claude Desktop app for your MCP server.
- Have your HO6 Skill and HO8 MCP server ready — this capstone composes your own pieces.
- Open a sample below and read its
README.md. - Follow its "Use it with your Claude.ai subscription" steps — add your Skill in Claude → Skills, add your MCP in Claude Desktop, and (optionally) turn on a Connector in Settings → Connectors.
- Paste the sample's ready-made example prompt into a new Claude chat and watch the agent orchestrate your MCP + your Skill end-to-end. Capture a run log/recording.
Each sample is copy-and-use: the Skill, the example prompt, and a real sample run are all filled in. Start from one, then swap in your own HO6 Skill and HO8 MCP.
| # | Sample | The agent orchestrates… |
|---|---|---|
| 01 | Content Creation Pipeline | your MCP pulls the trend → your Skill drafts the post → (optionally) publish |
| 02 | Lead Enrichment & Outreach | your MCP enriches a lead → your Skill drafts outreach → log it (optional CRM Connector) |
| 03 | Daily Ops Data Pipeline | your MCP fetches data → your Skill analyses it → agent files/notifies, on a schedule |
| 04 | Support Ticket Workflow | your MCP fetches the ticket → your Skill drafts the reply → agent updates the tracker |
| 05 | Bring Your Own | your own HO6 Skill + HO8 MCP composed into one agent for a real recurring job |
Open index.html in your browser for a clickable overview of all five.
Pick one sample (or bring your own recurring job), wire your HO6 Skill and your HO8 MCP under an orchestrating agent (optionally a Connector to deliver), run it end-to-end, and submit:
- the agent config + your HO6 Skill and HO8 MCP wired in (optionally a Connector),
- a run log/recording of it composing them end-to-end,
- a README and a flow diagram of the pieces it orchestrates.
Each sample's connector.md and mcp-config.json show the fuller manual setup for running unattended on a schedule. That path uses platform API keys and a terminal and is not required for the course.
HO10 · Peer-reviewed · ForgeMind AI Course · module M11 (Week 4)
The Pro plan has a usage limit that resets every few hours. A few habits make it stretch — and keep a mistake from burning your whole session:
- Use the example prompt in each sample's README — it's already written and tested. Don't reinvent it.
- One clear prompt beats lots of vague back-and-forth. Say what you want, with an example, in a single message.
- Start a new chat when you switch tasks. Long chats re-read every earlier message and use up your limit faster.
- Don't paste big files over and over. Paste once, then refer back to it.
- If something works, keep it. Tweak it rather than regenerate from scratch.
- Using Claude Code or Cowork? This repo's
CLAUDE.mdmakes Claude follow these same rules automatically, andSKILL.mdis a reusable "token-wise" skill.
If you do hit the limit, it resets after a few hours — nothing you've saved is lost.