Open-Athena / Buckler Lab, 2025-07-28
- Copilots
- LLM web-apps (ChatGPT, Claude, Gemini)
- Terminal-based agents (Claude Code)
- Editor-based agents (Cursor, Windsurf)
Open-Athena/ec2: self-terminating GHA EC2 runner
Built on:
- Open-Athena/start-aws-gha-runner:
- GitHub Action to start an AWS EC2 instance as a GitHub Actions runner
- Fork of omsf/start-aws-gha-runner
- Open-Athena/gha-runner:
- Library for building infrastructure provisioning GitHub Actions via Docker in Python
- Fork of omsf/gha-runner
runsascoded/awair: air quality sensor dashboard
- Demo: awair.runsascoded.com
- AWS Lambda updates Parquet file in S3 every 3mins (per Awair API rate limits)
- Vite+React webapp fetches and display data from S3 Parquet
runsascoded/watchy: Python library and CLI tool for fetching and tracking GitHub stargazers and followers
Demo use: ryan-williams/.watchy
- Terminal UI (TUI): editor agnostic, easy remote use
- Unfamiliar languages
- ryan-williams/hammerspoon-files: Lua configs for Hammerspoon window manager
- Stagnant for years
- Recently added multiple features, thanks to
claude
- ryan-williams/hammerspoon-files: Lua configs for Hammerspoon window manager
- Unfamiliar frameworks / tools
- e.g. AWS Lambda usage in runsascoded/awair
- Iterating on build/test errors
- Fetches/Reads docs/code from internet, uses tools
Things I've observed (not exhaustive!):
- Tends to make code monotonically larger + more verbose
- Can ignore
CLAUDE.mdinstructions- e.g.
npm run dev(attempts to run dev server, that never returns) - Hearsay: "Forbidden to use …"
- e.g.
- Excessive sycophancy / suggestibility
- Niche languages / frameworks
- Doesn't read all the code (e.g.
head/tailto slice 10-20 lines at a time)
-
An Image is Worth 32 Tokens for Reconstruction and Generation
-
-
ryan-williams/git-helpers: 1,616 Git aliases and scripts
