rememberlisted

Distill durable lessons into `.ultragoal/memory/`. Focus given by the user (may be empty — then review the whole session and any just-finished goal, especially its Decision journal and Verification log): <focus> $ARGUMENTS </focus> ## The two layers — never mix them Every memory file has two layers separated by `---`: - **Above the line — compiled truth.** The current understanding, rewritten freely as it improves. This is what future sessions read first. - **Below the line — evidence log.** Append-only, dated, never rewritten or deleted. One self-contained line per observation: `[YYYY-MM-DD S<n>] <what was observed and how>`. Synthesis can be wrong; evidence can't. When you update a compiled claim, the evidence that justified the old version stays in the log. Anyone can re-derive the truth from below the line — that's the point. ## Where things go Each file's header says what belongs in it and what doesn't — read it before filing. The short version: - `facts.md` — what is *true of this repo* (schemas, behaviors, invariants, tool quirks) - `patterns.md` — *reusable approaches* that worked here, with why they worked - `failures.md` — *dead ends*: what was tried, why it failed, what to do instead - `MEMORY.md` — the index and the fixed slots (commands, invariants, gotchas, hot files). Fill `[no data yet]` slots as you learn; keep one index line per entry. Its head is injected into every session — keep it tight. Unsure where something goes? File the evidence line in the