Omoi
An iOS app for the writing that fandom produces — letters, radio submissions, conversations worth keeping — with statistics to look back through.
tkana
I build iOS apps and command-line tools on my own time.
Projects
Subagent Insights
Claude Code leaves a transcript behind for every subagent run. This reads them and reports how the agents actually behave — deterministically, with no LLM call and no network access.
Writing
GitHub Actionsを棚卸ししたら、CIに残す理由があったのは3つだけだった
After moving heavy macOS jobs to local pre-push hooks, only three CI tasks turned out to truly need cloud execution: AI code review, reproducible release builds, and multi-OS test matrices.
Claude Codeの「仕様なので人がやって」を辿ったら、私が書かせた記録だった
A wrong premise about iOS Keychain behavior got written into my own Claude Code rules as "verified," and traced back to a measurement command that never measured what it claimed to.
Geminiの無料枠でCIのAIレビューを回していたら、レビューが走っていなかった
My CI's AI code review looked healthy while it silently fell back to a weaker model and swallowed rate-limit errors as "no issues found."
サブエージェントの通信簿を作ったら、自分の指示の穴が出てきた
A tool that scored my Claude Code subagents turned out to be measuring nothing, and the defects it hid say more about how I brief agents than about the agents.