claude-memory-kitlisted

# Claude Memory Kit v4 Persistent memory for Claude Code agents. Layered memory, agent-driven promotion, zero manual file editing. ## Two core invariants 1. **User only talks. Agent captures, proposes, writes.** Every architectural decision passes this test. 2. **Every memory entry carries a `[YYYY-MM-DD]` date tag.** This is what lets `/close-day` detect cross-session repetition and propose promotions. ## What's different from v3.2 - **Agent-driven promotion ritual** via `/close-day` (was: background `promote-patterns.py` detection + `flush.py` auto-flush, both killed as unreliable and invariant-violating) - **Multi-project isolation** via `projects/<name>/` — shared layers load always, per-project scope on demand - **`experiments/<name>-YYYYMMDD/`** sandbox layer next to `projects/` — different lifecycle, no direct promotion to rules - **Date-tagging promoted to a documented load-bearing invariant** (was implicit machinery, now stated) - **Killed:** `experiences/` staging layer, `promote-patterns.py` + `flush.py` background scripts, `playbooks/` and `<role>-guidance/` reference-skill seeds, the `/memory-audit` and `/memory-compile` operators, and the `knowledge/connections/` + `knowledge/meetings/` subdirs. The role-guidance pattern still works if you add it per-project (see `.kit/ARCHITECTURE.md` § "Adding role-guidance yourself"); the kit just doesn't ship templates. - **Default surface trimmed to two operators** (`/close-day` + `/tour`). The wiki-maintenance command