@mindrian_os/cli 1.13.1-beta.20 → 1.13.1-beta.24

package/CHANGELOG.md

@@ -1,3 +1,19 @@
+## [1.13.1-beta.24] - 2026-06-14
+
+### Added
+- Phase 150.9 (doctor drift-classes): `/mos:doctor --drift` opt-in flag adds two report-first drift classes on top of the install-cache classes. Class P (prose-vs-code) scans Larry's agent/skill/doc prose for claims that contradict shipped facts; report-only, never auto-edits. Class Q (GSD-record drift) shells out to `gsd-health` to surface ROADMAP gaps + missing SUMMARYs as drift findings. `lib/core/drift-baseline.cjs` writes per-folder + root `DRIFT.md` baselines (idempotent, traversal-guarded) so successive `--drift` runs diff against a stable anchor; `--drift --fix` heals only where safe. First production run caught 96 W007 (ROADMAP gaps) + 9 I001 (missing SUMMARYs) - the Fable FIX-12 drift set. Requirements DDC-01..08; phase gate `tests/run-all-150.9.sh` 6/6 (incl. Canon Part 8 floor + deadlock carve-out proofs). Local-only: zero Brain egress.
+- Phase 150.10 (systems-thinking meta-lens): `/mos:systems-thinking` promoted in place from a flat connector to an F-surface move-selector (M1 name the system -> M2 draw the loop -> M3 find the leverage point -> M4 trace the reverse salient -> M5 validate), GUIDED, with 3-layer local ranking and stage-aware filing. `lib/core/leverage-scan.cjs` is the M4 move: a Meadows 12-level leverage scan over `room.db` (reads via `navigation.cjs` only, ranked highest-leverage-first) wired into the selector as a Decision Gate, with the meta-lens chaining web (M4<->reverse-salient, M3<->find-analogies+research) wired in code. `systems-thinking-loop` registered as a ranked reach component in `lib/hmi/reach-component-map.json` (FROZEN-6 held: REACH_IDS still 6, byte-unchanged). Requirements ST-01..18; phase gate `tests/run-all-150.10.sh` green.
+
+### Changed
+- Brain teaching graph (production Neo4j): +12 generic-methodology nodes for the IRIS 2026 Session 2 lecture (`source_doc='iris-2026-session-2'`, all MERGE-based / idempotent) - the M1-M5 Method nodes + PREREQUISITE chain (M1->M2->M3->M4->M5, M4 FEEDS_INTO M5) + 4 Example nodes + trending-to-absurd Technique + the Leverage Point Local-Graph Excavation Method + cross-domain chaining edges (M4<->Reverse Salient, M3->Four Lenses, Excavation->M4/Leverage Points/Reverse Salient). Generic methodology only, zero user content (Canon Part 8 clean, orphan-scan 0). Re-ingestion is idempotent.
+
+## [1.13.1-beta.22] - 2026-06-13
+
+### Added
+- Conversation intake plumbing (quick 260612-pkb): the Stop and PreCompact hooks now capture Claude Code's `transcript_path`, parse the conversation offline (`scripts/transcript-ingest.cjs` -- user content as string, assistant text blocks only, drops thinking/tool_use/command-chrome, caps 4000 chars/fragment and 1000 fragments), and write real `user`/`assistant` fragments into the active room's `room.db` before summarizing. The dormant downstream readers (voice_log writer, `sessions.summary`, the RECENT SESSIONS resume block) light up unchanged. Fixes the "Brother Test" (reopen a room and Larry references the prior session) and Context Volatility failure modes 2 (session termination) and 3 (unfiled insights). `sessions.summary` is now a 3-5 sentence extractive summary over real turns instead of the `session ended` stub. Local-only (Canon Part 8): zero Brain egress.
+- Proactive Filing Offer doctrine (quick 260612-t29): `skills/room-proactive/SKILL.md` gains a "Proactive Filing Offer (Conversation Artifact Capture)" section -- when a conversation yields a keepable artifact (problem definition, competitive landscape, decision, plan/pilot, synthesis), Larry closes the turn with a Decision Gate F.1 selector mapped onto the existing canonical verbs (Run Methodology / Bank Opportunity / Synthesize / Defer / Free-Text), with precise triggers + anti-triggers (one offer per artifact, escape hatch honored) and decline-as-data capture (Canon Part 4). `skills/larry-personality/SKILL.md` anchors the behavior in conversation flow. No new verb, no new selector (Canon Part 7 reuse).
+- Persona override, identity-only (quick 260612-t2k): new `lib/core/persona-override.cjs` store (`~/.mindrian/persona-override.json`; get/set/clear/status CLI, taxonomy-validated, atomic write, local-only) plus a `readUserMd` seam (`lib/core/user-md-ops.cjs`) that returns a synthetic persona to all ~9 persona callers from one chokepoint while an override is active. The store lives outside the context window, so a navigator-declared persona survives maintenance commands like `/mos:doctor` -- the exact failure that collapsed the persona in QA. `commands/persona.md` documents the set/status/clear surface. The no-override path is byte-identical (proved by the 22/22 user-md-persona regression); `tests/test-persona-override.cjs` adds 5 assertions. Wires the previously-dead `detectPersonaUpdate` `user_override` case.
+
 ## [1.13.1-beta.20] - 2026-06-12
 
 ### Added

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindrian_os/cli",
-  "version": "1.13.1-beta.20",
+  "version": "1.13.1-beta.24",
   "description": "Install MindrianOS into Claude Code with one command -- `npx @mindrian_os/cli`. Ships the MindrianOS plugin (Larry + PWS methodology + Data Room) plus a setup/diagnostics CLI (install/doctor/update).",
   "scripts": {
     "mcp": "node bin/mindrian-mcp-server.cjs",