Latest Updates
Yes, Claude Code Can Generate Images — Calling Images 2.0 via OpenAI's Official Codex Plugin (+ 4 Pitfalls and a Delegation Design)
Claude Code can't generate images on its own, but delegate to Images 2.0 through OpenAI's official Codex plugin and it can. Here are the 4 pitfalls I hit while making it production-ready (dies in the background, sandbox path limits, CJK garbling, hangs via Skill) and the delegation design — parent/child role separation plus real verification.
Claude Code Routines × GitHub: Pitfalls and Workarounds
A breakdown of timeout limits and sandbox authentication constraints I hit when saving files via the GitHub MCP from Claude Code Routines, along with workarounds for each.
Enabling and Using Codex's experimental Hooks
How I discovered Codex's experimental Hooks feature, enabled it, and reverse-engineered its config structure and supported events by reading the source.
Hands-on with Claude Code's /loop Feature
Using Claude Code's /loop to define recurring jobs in natural language. Practical patterns for API and issue monitoring, gotchas around auto-compact and interval design, and how it interacts with Hooks.
Claude Code Agent Teams: Reusing Existing Skill and Agent Knowledge
An overview of Agent Teams, how it differs from traditional subagents, the current state of reusing existing skills and agent definitions, and what token consumption actually looks like.
Lost in the Middle — Prompt Design that Beats LLM Position Bias
LLMs tend to overlook the middle of long prompts (the Lost in the Middle problem). This post covers the cause and practical countermeasures — tail checklists, the sandwich strategy, XML-tag structuring, and more.