Chain JARVIS steps so it finishes a multi-step task on its own, recovering when a step fails.

Warm up

1. Warm up 01

   Take a task too big for one prompt

2. Warm up 02

   Break it into agent steps (plan-act-observe)

3. Warm up 03

   Run it; watch it branch

4. Warm up 04

   Add a recovery path for the failing step

Challenge

1. Challenge 01

   Decompose a 4-step task into agent steps

2. Challenge 02

   Design a fallback for a failing step

Take further

1. Take further 01

   JARVIS completes a multi-step task unattended

2. Take further 02

   A research agent (search-read-summarise)

3. Take further 03

   An agent that retries intelligently

Controlled practice