Run the Framework End-to-End

In this exercise you run the AI Workflow Framework end-to-end yourself, live in class. Starting from a pre-built Workflow Requirements, you invoke /design, /build, /test, and /run on a Competitive Intelligence workflow — shipping a competitor-research skill and competitor-brief agent that produce a structured brief on a real competitor. In a follow-up session, you’ll evolve that workflow into a self-improving system.

Lesson: Ship a Workflow Using the AI Workflow Framework

Run the framework end-to-end yourself in Cowork (or Claude Code — same slash commands). Starting from a pre-built Workflow Requirements, you invoke /design, /build, /test, and /run — generating a competitor-research skill and competitor-brief agent from your approved spec, then running the workflow on a real competitor. Watch a structured context file get produced on the first run.

Hands-on assignment

Starting point: a pre-built Workflow Requirements (download below). Ending point: a shipped skill + agent producing a brief on a real competitor.

1. /design — Turn the requirements into an approved Design Spec (plan mode, collaborative). See the Design step docs.
2. /build — Generate the competitor-research skill and competitor-brief agent from your spec. See the Build step docs.
3. /test — Validate the building blocks before trusting them with real input. See the Test step docs.
4. /run — Invoke the workflow on a real competitor; watch knowledge/competitors/{name}.md emerge. See the Run step docs.

Download the Workflow Requirements

The Step 2 (Deconstruct) artifact is pre-built so we can spend class time running the rest of the framework on it.

Download the Competitive Intelligence — Workflow Requirements (.md)

Set up your workspace before class

Do these steps in the folder where you run Cowork or Claude Code.

Step 1 — get a registry/ folder. Pick whichever is true for you:

• Starting fresh, no workspace yet: instantiate the template repo.
  1. Open the AI Registry template repository: github.com/jamesgray-ai/ai-registry-template (or ask your AI assistant: “Where’s the AI Registry template repo?”).
  2. Click the green Use this template button near the top of the page, then Create a new repository.
  3. Clone the new repository to your computer (or open it directly in Cowork) — it arrives with an empty registry/ folder already in place.
• Already have a workspace: open it in your AI assistant and say “Set up my AI registry.” — it creates the registry/ folder for you.

Step 2 — create the workflow folder and save the download. Inside your workspace, create a folder named outputs, and inside it a folder named competitive-intelligence-brief. Save the download as requirements.md inside that folder, so the full path is outputs/competitive-intelligence-brief/requirements.md. (Rename the downloaded file — the framework skills look for requirements.md.)

Step 3 — hand-type the first Workflow node. In registry/workflows/, create a file named competitive-intelligence-brief.md and paste in exactly this. Easiest path: on github.com, open your registry repository, click Add file → Create new file, paste registry/workflows/competitive-intelligence-brief.md as the filename, paste the block below as the content, then Commit. (Or, if you’re working locally, ask your AI assistant to create the file for you with this exact content — same result either way.)

---
type: Workflow
title: "Competitive Intelligence Brief"
description: "Research a competitor's recent moves and produce a structured brief plus an updated knowledge file."
generated: { by: process:deconstruct, at: 2026-07-06 }
status: under-development
definition_type: step-driven
execution_mode: manual
autonomy: guided
trigger: "manual"
stale_after: 2026-10-06
---
# Competitive Intelligence Brief

Research a competitor's recent moves and produce a structured brief plus an
updated knowledge file. Triggered manually on demand, or scheduled per competitor.

# Artifacts

- [Requirements](outputs/competitive-intelligence-brief/requirements.md)

The node is the small file every framework step reads first; your registry/ is the inventory. When class starts, you’ll run /design and it will pick up right where the pre-built requirements leave off.

What you’ll walk away with

• A competitor-research skill and a competitor-brief agent installed in your workspace, ready to invoke any time
• A populated knowledge/competitors/{name}.md file on a real competitor — the seed for the self-improvement work later in the course
• First-hand reps on every framework step as a chainable slash command (/analyze, /deconstruct, /design, /build, /test, /run, /improve)

Lesson: Recognize When a Workflow Needs to Improve

A brief, conceptual preview of Step 7 of the framework — Improve — anchored to the workflow you just ran. Identify the three signals that tell you a running workflow needs to evolve: hypotheses that never confirm, edge cases where the schema breaks, and drift as your priorities evolve. Improve is the step that never ends — and in the follow-up session, you’ll actually do the iteration.

Lesson: Evolve Your Workflow Into a Self-Improving System

Take the workflow you just shipped and evolve it into a system that gets sharper every time it runs. Apply Karpathy’s “LLM Wiki” pattern to its structured context file — structure it with Rules / Facts / Hypotheses sections, implement an ingest / query / lint learning loop, and schedule the workflow to run daily without you. By session end, your workflow’s outputs feed a self-improving knowledge base of competitor intelligence — Step 7 (Improve) in action.