2026-04-12 Thinking traces and chezmoi fixes
12:16 PM I have finally fixed something via chezmoi that has been bugging me. My AGENTS.md and CLAUDE.md keep diverging on the support files across my multiple machines. The attempt I did today was to codify it into the chezmoi repo, rather than the user-level CLAUDE.md file which I feel is getting overloaded. I can already feel that parts of the user instructions get ignored often. Maybe it was because references are broken. I don’t know. Let me see if this fix fixes that.
4:20 PM I didn’t know this until I saw it from @emollick: thinking traces in LLMs are always summaries of the internal reasoning — the actual internal chain of thought is a trade secret. That doesn’t mean they aren’t useful when summarized, especially when those summaries include the tools used by the model and web searches it made. Makes me think about how much of what we see as “transparency” in AI reasoning is itself a curated output.
