Not philosophy — artifacts. Slash commands, how-to guides, framework rules, and a working req/reply protocol, all built through months of daily use.
Claude Code itself doesn't have an opinion about this. It's a tool — powerful, but it assumes you'll remember what you asked for ten minutes ago. In practice, across long sessions and dozens of parallel projects, the human is the one who forgets, not the AI.
T00's req/reply protocol exists specifically because of that: every substantive answer gets written to a persistent log instead of only spoken into a chat window that scrolls away. It's not a feature for the AI's benefit — it's a feature for mine.
The instinct that's hardest to teach an AI is the one experienced engineers build over decades: when something looks safe to simplify, and when it only looks that way. Here's a real example from this site's own build log, not a hypothetical.
The ask: a dropdown menu needed "one click instead of two" to open.
The easy answer: add a click handler. Looked done. Wasn't — the real cause was a CSS gap the mouse had to cross, silently breaking the hover state.
The judgment call: don't patch the symptom, find why it actually happens. Closing the gap fixed it for good, with no JavaScript required — the kind of root-cause instinct that mostly comes from having shipped, and debugged, a lot of software before.
That instinct is what T00 tries to encode as it goes: not just what was built, but why — so the same mistake doesn't get "fixed" back into existence six months later. Curious how that's tracked? See the T00 Knowledge Base →
The T00 Knowledge Base breaks down the eight concrete concepts that produced all this — the req/reply protocol, trust levels, persistent memory, slash commands, skills, mapping requests, design docs, and honest status reporting.