Sterling G. Baird sgbaird

Bring a new Raspberry Pi onto the lab tailnet

A reusable Claude Code prompt. Start a session in the repo the Pi belongs to, fill in the Inputs block, and paste everything from the horizontal rule down.

Distilled from how the vertical-cloud-lab Pis were actually brought up: byu-vcl (CLAUDE.md, its Tailscale → Raspberry Pi connection section, and .github/workflows/claude.yml), powder-doser, tensegrity-optimization,

Pin a repo to a specific Claude Code account

If you have more than one Claude account on a machine, Claude Code uses whichever one the default config dir is logged into. That's fine until you have repos that should bill to different accounts. This pins a repo to one account, deterministically.

How it works

Claude Code keeps all auth/session state in one directory, ~/.claude by default. The CLAUDE_CONFIG_DIR env var overrides that path. Point it at a second directory

@claude+opus:max download `` (use the Pi to do this) and extract the transcript and get screenshots of every place where there's some actionable feedback point from the transcript (and make sure to correct the transcript for intent since there's probably voice-to-text mishaps), and turn that into a full md document that you commit to this branch (there will be lots of screenshot images, so these images will need to be put in a subdir). Also grab comments on that video, if any, and inlay each comment line into the corresponding place in the video. Also make sure to include timestamps for where the screenshots were taken (and you might need to take multiple screenshots in quick succession, since sometimes comments made aloud correspond to where the mouse pointer is at exactly that moment). This is a high effort task and will require a good amount of image analysis. Every two minutes, check for follow-up comments that specifically relate to this video and address them. Your job is to implement the f

Development Practices

Start with minimal, lean implementations focused on proof-of-concept
Avoid creating new files until asked
Avoid implementing things from scratch
Avoid defensive error handling for hypothetical situations For example, rather than using:
```
...
try:
```

	#!/usr/bin/env python3
	"""Re-mint the GitHub App token an agent session runs on, mid-session.

	Why this exists: claude-code-action mints a GitHub App installation token once,
	at the start of the Run Claude Code step, and installation tokens live exactly
	one hour. A session that outlives the hour keeps running, but git push, gh, and
	the MCP comment tool all start failing with 401. The action deliberately strips
	ACTIONS_ID_TOKEN_REQUEST_URL and ACTIONS_ID_TOKEN_REQUEST_TOKEN from the
	session's environment (base-action/src/parse-sdk-options.ts), so the session
	cannot re-run the exchange from its own env. Those two values are still

	name: Claude Code Review

	on:
	pull_request:
	types: [opened, ready_for_review, reopened]
	# Optional: Only run on specific file changes
	# paths:
	# - "src/*/.ts"
	# - "src/*/.tsx"
	# - "src/*/.js"

	name: Claude Code

	on:
	issue_comment:
	types: [created]
	pull_request_review_comment:
	types: [created]
	issues:
	types: [opened, assigned]
	pull_request_review:

	import numpy as np, matplotlib
	matplotlib.use("Agg")
	import matplotlib.pyplot as plt, io, os, math
	import PIL.Image as Image
	import imageio.v2 as imageio

	# Parameters
	L = 1.0
	D = 1.0
	Nx = 201

	from opentrons import protocol_api

	metadata = {
	"apiLevel": "2.16",
	"protocolName": "Flex Gripper Wellplate Demo",
	"description": "Demo showing how to move a wellplate using the Flex gripper"
	}

	def run(protocol: protocol_api.ProtocolContext):
	# Load a 96-well plate in deck slot D1

	name: "Copilot Setup Steps"
	on:
	workflow_dispatch:
	push:
	paths: [.github/workflows/copilot-setup-steps.yml]
	pull_request:
	paths: [.github/workflows/copilot-setup-steps.yml]

	jobs:
	copilot-setup-steps: