@korrlabs/mnemospi 4.1.0 → 4.3.0

package/README.md

@@ -114,7 +114,7 @@ One local server — and a connected agent harness gets the full memory stack.
 | **Lifecycle hooks** | `pre_llm_call` context injection, `on_session_start`, `post_tool_call` auto-compression of tool outputs |
 | **Publication v3.0.0** | Entries visible immediately after save, background refinement with seamless swap, quarantine with neutral retraction |
 | **Self-protection** | Injection / secret detectors on input and publication, every output scanned, full per-entry audit |
-| **Auto-pipeline** | Background processor: clustering, deduplication, quality gate, publication |
+| **Auto-pipeline** | Background processor: clustering, deduplication, quality gate, publication. Entries awaiting refinement sit at `pipeline_state=pending` until the processor runs — in CLI-only deployments (no daemon) start it with `mnemos processor start`; `mnemos doctor` reports the pending-queue depth |
 
 Autonomy for an arbitrary harness and LLM-driven enrichment are partial — the
 full, honest map lives in [docs/en/features.md](docs/en/features.md).
@@ -142,6 +142,15 @@ control surfaces, and a storage layer you can read with your own eyes.
 SQLite for metadata, a local numpy + SQLite vector index for recall, and an Obsidian-compatible vault
 for the humans in the loop.
 
+**Where this is going.** The shipped, measured rung is *stores and finds*. The next rungs —
+*collapse with checks* (session → project → cross-project synthesis: automatic by default, but never
+unconditional in authority — every derivative traces back to its sources, nothing pins without an
+operator) and, later, *builds understanding* — open only as pre-registered experiments prove them
+([ADR-0025](docs/project/adr/0025-memory-meta-level-lanes.md)). The invariant on that road is
+**zero silent losses**: a fact is either retained, or its loss is visible. And the shape of the
+ambition is a *nervous system, not a conductor* — memory that surfaces the right thing at the right
+moment, never one that conducts the agent.
+
 ---
 
 ## 🤝 Connect any harness
@@ -303,3 +312,27 @@ Apache-2.0 — see [LICENSE](LICENSE) and [NOTICE](NOTICE). Source: [github.com/
 
 Contributions are welcome — [CONTRIBUTING.md](CONTRIBUTING.md) has the development setup, the branch
 and commit conventions, and the quality gate a change must pass.
+
+---
+
+## 🛠 CI / Релизы — кластерный конвейер `release-pipeline`
+
+Этот проект подключён к общему кластерному конвейеру релизов
+([Korrnals/release-pipeline](https://github.com/Korrnals/release-pipeline),
+K3s `abyss-ai-agent`, namespace `release-pipeline`). GitHub Actions не
+используется (биллинг аккаунта заблокирован) — конвейер и есть штатный путь
+релизов. Релизный артефакт подписывается трёхслойно: SHA256 → SBOM → cosign → GPG.
+
+**Релиз новой версии:**
+1. `VERSION` → релизный коммит (конвенция репо) → тег `vX.Y.Z` → push.
+2. Обновить версию проекта в `projects[]` файла `~/.cache/release-pipeline-values.yaml` и применить:
+   ```bash
+   helm upgrade --install release-pipeline \
+     ~/LABs/Projects/Project-Umbra/release-pipeline/chart/release-pipeline \
+     --kube-context abyss-ai-agent -n release-pipeline \
+     -f ~/.cache/release-pipeline-values.yaml
+   ```
+3. Наблюдение: `kubectl --context abyss-ai-agent -n release-pipeline get jobs`,
+   логи: `kubectl ... logs -f job/release-<proj>-<ver>`.
+
+Подробности (типы проектов, kaniko-контейнеры, teardown): README конвейера.

package/README.ru.md

@@ -142,6 +142,15 @@ mnemos search "помнит между сессиями"
 SQLite для метаданных, локальный векторный индекс на numpy + SQLite для recall и Obsidian-совместимый
 vault для людей в контуре.
 
+**Куда это движется.** Отгруженная и измеренная ступень — *хранит и находит*. Следующие ступени —
+*свёртка с чеками* (синтез сессия → проект → кросс-проект: автоматическая по умолчанию, но никогда
+безусловная по авторитету — каждый итог возводим к исходникам, ничто не пиннится без оператора) и
+позже *строит понимание* — открываются только по мере подтверждения предрегистрированными
+экспериментами ([ADR-0025](docs/project/adr/0025-memory-meta-level-lanes.md)). Инвариант этого
+пути — **ноль тихих потерь**: факт либо сохранён, либо потеря видна. А форма амбиции — *нервная
+система, а не дирижёр*: память, которая поднимает нужное в нужный момент, но никогда не дирижирует
+агентом.
+
 ---
 
 ## 🤝 Подключение любого харнеса

package/integrations/skills/mnemos-context-lifecycle.md

@@ -28,6 +28,10 @@ rewrite lifecycle.
 - **Tool output compression (hooks.post_tool_call)** — a tool returned a
   huge output you want substituted by a zero-loss CCR marker, with
   provenance stamped for strict marker validation.
+- **Re-assembly (mid-session)** — only when memory changed materially (a
+  new decision, a rewritten context block that must surface). The
+  assembled block is prefix-stable by design; gratuitous re-assembly
+  rewrites the prefix and discards the cache the session was building.
 
 ## STEPS
 
@@ -92,6 +96,22 @@ rewrite lifecycle.
    project/agent/session/supersedes/content), so re-delivery cannot
    duplicate writes.
 
+6. **Keep assembly cache-friendly** (provider-agnostic cache discipline):
+
+   - Assemble the context once per session, or on material memory
+     change; mid-session re-assembly is reserved for significant
+     changes only.
+   - The assembled mnemos text is per-call conversation content (tail),
+     NOT a standing prefix — never inject mnemos blocks into the system
+     prompt.
+   - Do not change the MCP tool set or tool schemas mid-session; a tool
+     change invalidates the provider's cache for the entire session.
+   - Prefer append-over-rewrite on compaction: append the new summary
+     after the stable prefix rather than rewriting the prefix.
+   - Align system prompts once, at assembly time
+     (`mnemos_align_prefix`); dynamic values (timestamps, counters,
+     volatile state) go to the tail, never the prefix.
+
 ## DISCIPLINE
 
 - **Never assemble context by hand.** The pipeline stages (context filter,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@korrlabs/mnemospi",
-  "version": "4.1.0",
+  "version": "4.3.0",
   "description": "Mnemos memory & knowledge server — Pi extension. Spawns `mnemos mcp-server` over stdio and exposes every mnemos_* tool as a native Pi tool, plus the mnemos skill pack.",
   "type": "module",
   "license": "Apache-2.0",