rememberlisted

# remember The sasu guy is never told the same thing twice. This skill turns a lesson into a landed asset: a documented fact, a machine-checked invariant, or a regression test. A lesson that only becomes prose is not learned; it must end up somewhere that a machine or a future session is forced to encounter. Match the user's language by default. ## The Landing Model Every lesson gets classified twice, then landed: | Kind | What it is | Landing | Enforced by | | --- | --- | --- | --- | | fact | something agents must know (build quirks, gotchas, commands) | body in `docs/**`, one index line in AGENTS.md | auto-loaded context | | invariant | "when X changes, Y must hold" | `agents/rules/invariants/<ID>.md` via `rules add` | `ship` pre-push gate + implement plan injection | | regression | "a test would have caught this" | a real test in the project's own suite | the project's CI / verification plan | | Route | Where it goes | | --- | --- | | project | the target project, per the table above | | user | the user's shared reference files (`~/.claude/CLAUDE.md`, `~/.codex/AGENTS.md`), only with explicit confirmation | | harness | a proposed diff to the engineering-harness repo (SKILL.md or scripts); never applied without confirmation | ## Inputs - An explicit lesson from the user in conversation. - `rememberSuggestions` from a `finalize` receipt, and the `deviations` array in the latest run's `state.json`. - A just-finished incident: name the cause before it evaporates. ## M