Skip to content

Instantly share code, notes, and snippets.

View keithloweryamd's full-sized avatar

keithloweryamd

5 followers · 0 following
View GitHub Profile
@keithloweryamd
keithloweryamd / claude-vs-codex-bespoke-agent-module-study.md
Created April 28, 2026 22:15
Claude vs Codex bespoke agent JavaScript module trace study

Claude vs. Codex Bespoke Agent JavaScript Module Trace Study

Date: 2026-04-28

This study compares two recent captured coding-agent runs that were asked to build a browser-based JavaScript module for creating bespoke AI agents. The requested module needed runtime selection between Anthropic Messages API and OpenAI/Codex Responses API, developer-defined JavaScript tools, skills, context management and compaction, an asynchronous event-emitting agent loop, browser and Node.js usability, unit tests, documentation, and exactly one single-page web app demo.

Final output directories:

  • Claude output: /Users/keithlowery/horserace/claude
  • Codex output: /Users/keithlowery/horserace/codex
@keithloweryamd
keithloweryamd / claude-vs-codex-pacman-trace-study.md
Created April 28, 2026 21:05
Claude vs Codex Pac-Man game-engine trace study

Claude vs. Codex Pac-Man Game Engine Trace Study

Date: 2026-04-28

This study compares two captured coding-agent runs that were asked to build a browser-playable Pac-Man game on top of a generalized game engine. The comparison combines trace analytics from AgentSnoop with filesystem inspection of the final outputs:

  • Claude output: /Users/keithlowery/horserace/claude
  • Codex output: /Users/keithlowery/horserace/codex
  • Codex trace group: heuristic:b4e3833b, Workspace: workspace
  • Claude trace group: heuristic:3dce4982, Plan: plans/hashed-purring-rivest.md