Codex already writes a local JSONL transcript for every session. A stop hook can use that transcript to append each prompt and response to Markdown, giving you searchable Obsidian notes without asking the model to summarize anything...
These are my learning notes and reflections while reading "AI Engineering: Building Applications with Foundation Models" by Chip Huyen. I plan to update this periodically as I progress through the book...
One of the most powerful features in Claude Code is the ability to create **custom slash commands**. Instead of repeatedly typing long or error-prone commands, you can encapsulate workflows into a reusable command...