Skip to content
Claude Code track

Use Claude Code when you want AI to work inside the project folder.

This track is for attendees who can use a terminal. Start with planning, keep permissions visible, and switch to a faster model once the build steps are specific.

Conference Order

Follow the pages in this order during the event.

1. Guide 2. Simple Setup 3. Models 4. Plan Mode 5. Pod Roles 6. Prompts 7. Claude Code 8. Codex 9. Hermes 10. Demo Prep 11. Support
Setup

Open the project folder

Work from the pod repo or local folder. Do not start in a random downloads folder.

Start Claude Code

Run Claude Code and confirm it can read the current folder. If install or login fails, move to the Hermes/fallback track so the pod can keep moving.

Use plan mode first

Ask Claude to inspect the folder and make a plan before it edits files.

Switch models after planning

Use the strongest available model for the plan and a faster execution model for narrow edits.

Auto mode with care

Auto or accept-edits mode is useful after the scope is clear because it reduces approval prompts. Keep it bounded to one approved step, one page, one feature, or one testable artifact at a time, and turn it off before broad cleanup, deletion, or anything involving credentials.

Move blocked users to fallback roles

If login, install, or permissions take more than ten minutes, the attendee should become researcher, tester, writer, or presenter while a builder continues.

Copy This
Read this folder and explain what is here.
Then make a plan for the smallest working version we can finish today.
Do not edit files until we approve the plan.
Switch to auto mode for the approved next step only.
Execute step 1, show what changed, and stop for review before step 2.
Now execute only step 1 of the approved plan.
Keep the change small, tell me which files changed, and run the basic check when done.
Safety Check

Good auto-mode work

Small file edits, formatting, one page, one component, one test, or one documented setup step.

Stop and ask first

Deleting files, changing payments, changing deployments, committing to GitHub, or touching secrets and tokens.

If the model drifts

Turn off auto mode, ask for a status summary, and switch back to planning with the strongest available model.