Workshop field guide · 93 minutes
Watch the loop become a system.
A timestamped companion to Joel Hooks’ live build: eight architectural turns, source-checked pull quotes, the canonical workshop curriculum, and prompts you can paste into your own agent.
Watch a chapter, open its matching lesson, copy one prompt, and save the requested proof before moving on. The video shows a live build; the lessons provide the repeatable public path.
Chapter map
Chapter 01
Workshop setup and the looping resolver
Define Loopcraft, start the Docker/Herdr/Pi rig, and frame the issue-progress loop.
Open the matching lesson →Chapter 02
XState, Effect, and owning the Pi harness
Choose explicit statecharts and typed effects instead of a free-form agent script.
Open the matching lesson →Chapter 03
Reviewing the scaffold and architecture
Inspect what the agent generated and separate useful structure from assumptions.
Open the matching lesson →Chapter 04
Planning before prompting
Turn a loose vision into constraints, stop rules, and a buildable sequence.
Open the matching lesson →Chapter 05
Daemon architecture and agent orchestration
Model durable ownership, events, receipts, and bounded handoffs between agents.
Open the matching lesson →Chapter 06
Compaction, tests, and enforcement
Watch the build under type, lint, test, and context-pressure constraints.
Open the matching lesson →Chapter 07
The Pi monitor and Lakebed UI
Project runtime facts into an operator surface without making the UI the domain model.
Open the matching lesson →Chapter 08
Running the loop end to end
Exercise the issue loop, observe failures, and identify the next hardening steps.
Open the matching lesson →Source-checked pull quotes
Short candidates aligned against the recording captions and retained only when the wording and timestamp agree. All quotes are Joel Hooks.
“Succeed over time and be something that you can rely on.”
“If you close the terminal window, your session persists.”
“Craft is something that I consider in terms of craft versus slop.”
“This is something that actually goes over the top of your repositories in GitHub.”
“Type safety is something that I think is incredibly important.”
“As a human, when I was typing all my code, I would have been super annoyed with it.”
“What I see in my mind’s eye is a statechart or a state machine.”
“An agent that autonomously loops over issues wherever they may be.”
“An agent that you control, that you have built from scratch.”
“I really love Pi as a prototyping tool.”
Copy/paste prompt lab
These workshop-inspired prompts use placeholders and stop before destructive or externally visible actions.
1. Define the loop before code
Copy vision prompt
Use the repository's grill-with-docs workflow to define a bounded local issue-progress loop before writing product code.
Read {{REPO_RULE_FILES}} and ask one question at a time. Capture the goal, files the loop may edit, approval points, non-goals, ownership rules, and stop conditions in {{VISION_FILE}}.
Do not create issue events, checks, UI code, background workers, external tracker integrations, or product code yet. End with a receipt listing files changed and confirming the work stopped at the agreed boundary.2. Turn the loop into an explicit machine
Copy statechart prompt
Design an explicit statechart for an autonomous issue-progress loop.
Inputs: {{ISSUE_EVENT_SCHEMA}}, {{APPROVAL_RULES}}, {{STOP_RULES}}, and {{RETRY_POLICY}}.
Return states, events, guards, actions, invoked actors, terminal outcomes, recovery paths, and the receipt emitted at each externally visible transition. Mark every destructive or external action as approval-required. Do not implement code until I confirm the statechart.3. Audit the evidence, not the vibes
Copy evidence prompt
Review this agent-loop change as an evidence auditor.
Compare {{VISION_FILE}}, {{STATE_MACHINE}}, {{EVENT_FIXTURES}}, {{TEST_OUTPUT}}, and {{RECEIPTS}}. Identify unsupported claims, hidden state, missing stop paths, unsafe retries, and places where the UI diverges from the event source of truth.
Return: verified facts, gaps, risk-ranked fixes, and the smallest next test. Do not modify files or contact external systems without confirmation.Resource shelf
Evidence policy
- The eight public lessons and repository files are the canonical repeatable workshop path.
- Video observations are separated from inference and checked against source captions or frames before publication.
- No credentials, callback URLs, private machine paths, account details, or raw transcripts are published.
- Version numbers and provider login flows shown in the recording may drift; verify current docs before use.