@mindrian_os/cli 1.15.0-beta.5 → 1.15.0-beta.9

package/CHANGELOG.md

@@ -1,3 +1,32 @@
+## [1.15.0-beta.9] - 2026-06-28
+
+### Added
+- **Phase 187 - Statusline navigator cockpit.** The statusline is rebuilt to serve the NAVIGATOR (not the operator), per the LOCKED `docs/STATUSLINE-CONTRACT.md` (Phase 121.5 co-design rule). Four tiers in a hierarchy: (1) trust metadata (Mindrian glyph + Voice Signature glyph + Brain), (2) orientation + room-health, (3) `Next: <move>` action cue, (4) `Ctx <n>%` risk. Color is carried by EMOJI GLYPHS (host-independent; this host strips ANSI). REORDER-AT-CLIFF: at >=80% context the line promotes "file this insight to the room before it compacts" to the hero slot; post-update drift promotes `-> run /mos:doctor --fix`. NORMATIVE anti-Dealer invariant INV-SL-1..5 (success metric = % of exposures that lead to a real advancing action, never time-on-line). `lib/statusline/cockpit-renderer.cjs` (pure) + `cockpit-signals.cjs` + `cockpit-telemetry.cjs` (LOCAL INV-SL-2 hook) wired into `scripts/context-monitor`; the two-row block is preserved. test-statusline-cockpit-187 14 tests/117 assertions; run-all-187 2/2.
+- **Phase 184 - READER decide-time projection offer (navigator-authority override).** `decide()` gains the projection read it lacked: it ranks capabilities from the LOCAL orchestration projection + connector registry + recipe-maps and surfaces them as Shape F Decision-Gate OPTION CONTENT. A third READER, never a firer (R4 structural guard: `decide()` has no code path to runChain/act-command, proven by test). `lib/core/reader/decide-projection-reader.cjs` (READER-01..04 + R2 projection-correctness gate + R3 ambient latency/context budget) + `ab-harness.cjs` (R1 A/B). run-all-184 2/2 (39+23 assertions); run-all-144 5/5 (no decide() regression).
+- **Phase 185 - DRIFT runtime reachability.** `doctor --drift` now FAILS when a WIRED capability is unreachable by `decide()` at runtime (previously merge-time marking only; CIRS R7). A capability is unreachable when it is WIRED in the connector registry and reader-eligible (command/agent) but the Phase-184 reader's deterministic ranker emits no candidate for its projection node. Calibrated GREEN on shipped data: 85 WIRED command/agent capabilities all reader-emitted, 5 WIRED skills correctly scoped out. Additive Class R in `scripts/doctor.cjs` + `lib/core/drift-runtime-reachability.cjs`; non-zero exit scoped so real `--drift`/`--all` stay green. run-all-185 1/1 (11 assertions); run-all-150.9 6/6 (no doctor regression).
+
+### Fixed
+- **Phase 182.1 - SIGNAL voice-glyph repair (dogfood fix).** Dogfooding v1.15.0-beta.7 found the Phase 182 Voice Signature was DARK at runtime (the doctrine lived only in the skill, not the always-loaded agent body, and the test asserted the doctrine string existed but never that a turn emitted a mark) AND that neither a bracketed color-name word nor ANSI escape codes render color on the navigator's host (ANSI is stripped to literal text). Fix delivers the 5 De Stijl Mondrian primaries as colored EMOJI GLYPHS (host-independent, verified): blue=building, red=challenging, yellow=contradiction, black=gate, white=invisibility. Same 5-primary palette; only the delivery mechanism moved from word/ANSI to glyph. Touches `agents/larry-extended.md` (the always-loaded body), `lib/hmi/voice-color-mark.cjs` (glyph-aware detector), both voice SKILL doctrines, and `lib/core/nav-dial.cjs` (the dial-TUI gains a leading glyph so its dead-ANSI color survives). Frozen MAX_K=3 / DIAL_REACH_K=6 / 0.70-0.15 / 6-reach bank UNCHANGED. voice-mark 106/106, dial PASS, run-all-182 5/5.
+
+### Known follow-ons
+- **184 R1 live A/B remains a NAMED DEBT.** Phase 184 was deferred as evidence-blocked; the navigator overrode the deferral on authority (the entry-20 pattern). The deferral reason does not vanish: R1's live grounded-vs-ungrounded A/B needs a real navigator reaching the gate (the same evidence METER came back missing, `subject_class=unknown`), so it reads `uninstrumented` today and a maintainer reading does not clear it. The CODE ships green; the live reading is recorded as a debt, not fabricated.
+- **187 host signals not exposed on the hot path:** `/mos:doctor` does not yet write a room-health status cache (cockpit reads `~/.mindrian/room-health.json`, defaults sound); the Voice-glyph WRITE side (a turn-capture hook) is unwired (glyph read from a side-channel); next-move uses the JTBD proxy. Post-update drift IS wired.
+- The entry-31 self-binding clause stays UNCLEARED (no navigator-class two-gauge reading yet); Appendix D entry 32 remains parked by design.
+
+## [1.15.0-beta.7] - 2026-06-27
+
+### Added
+- **Phase 183 - METER: the welded two-gauge meter (the milestone keystone).** LOCAL telemetry over the Part 9 navigation.cjs chokepoint. Adds `gate_reached` to the frozen EVENT_TYPES set (the single sanctioned additive, 86 to 87) and a single surface-shared emit beside the live reach_presented loop. Gauge 1 (invocation density) reads reach_presented + gate_reached; Gauge 2 (transfer) is three named-debt proxies (reject-reason-capture-rate, insight-to-validated-decision latency to status_promoted, independence-trend), each labelled named-debt, never a real transfer delta (Canon Part 5). `readTwoGauge` is WELDED: it returns the density+transfer pair or throws (no bare-density export), the two-directional regression guard names both volume-up-quality-flat (the Dealer quadrant) and quality-up-by-starving-volume. Two navigator-review corrections built in: a `subject_class` stamp (maintainer | navigator | unknown, derived read-time, Part-8-clean) where only `navigator` clears the entry-31 self-binding clause, and a `transfer_uninstrumented` third state so an empty substrate reads as unmeasured, never as flat. run-all-183.sh 8/8.
+- **Phase 182 - SIGNAL: the Part 12 Voice Signature in the CLI.** Every Larry CLI turn wears one De Stijl color mark naming the pedagogical move (blue=building, red=challenging, yellow=contradiction, black=gate, white=invisibility); a turn with no mark is legible as the native host. `lib/hmi/voice-color-mark.cjs` (detectVoiceMark) + doctrine on both voice SKILL surfaces, anchored to the 5 existing palette primaries (no new color minted). SIGNAL-01 verifies the Phase 179 GA-4 F.7 interceptor + R15 render-coverage gate still pass (lean, not rebuilt). Enforcement is the declared convention + a missing-mark drift test, not a per-token runtime recolor (the honest named residual). run-all-182.sh 5/5.
+- **Phase 186 - CORPUS: stats hygiene.** One generated single-source-of-truth artifact (docs/CORPUS-STATS.generated.md + json) from a committed Brain-free generator, plus a --check tripwire (wired into pre-commit + release.sh Step 2.4) that scans LIVE fact surfaces and skips historical provenance via a documented excludedRegion (the canon Appendix D dated entries, version-history rows, and .planning/ dated artifacts stay frozen). LIVE corpus literals repointed to 27,904 nodes / 177 frameworks / 12,485 Pinecone vectors. run-all-186.sh 3/3.
+
+### Changed
+- Canon v1.18 to v1.19 (Appendix D entry 31, the welded two-gauge headline metric + the Hooked-gate retirement + the self-binding clause) landed earlier in this milestone; METER is the instrument that entry 31 requires.
+
+### Known follow-ons
+- METER first reading is `subject_class: unknown` / `transfer_uninstrumented` (newborn instrument, no live-navigator gate-reach observed yet). The entry-31 self-binding clause stays UNCLEARED until a navigator-class reading exists; Appendix D entry 32 and the ProblemType freeze remain parked by design. Honest state: instrument live, awaiting a navigator.
+- Phase 184 READER + Phase 185 DRIFT deferred: READER is conditional on METER confirming a gate subject (unmet), and DRIFT depends on READER.
+
 ## [1.15.0-beta.5] - 2026-06-25
 
 ### Added

package/README.md

@@ -63,7 +63,7 @@ A raw AI is brilliant for one turn and forgets the next. Ask it the same venture
 MindrianOS adds the four things a raw model structurally cannot be:
 
 - **It remembers.** Your venture lives in a room that persists across every session. The contradiction you found three weeks ago is still there, still checking today's input.
-- **It knows when.** Not just which framework exists, but which one fits this stage and this problem, chained in the right order, calibrated from decades of teaching.
+- **It knows when.** The moat was never the frameworks; anyone can list those. It is knowing which one you need at the stage you are in, and reaching for it the moment you are stuck, so you never have to know its name. When to use which, in what order, is the timing that decades of teaching calibrate.
 - **It pushes back.** It blocks you when the evidence is thin and surfaces the conflict between your pricing and your market. A co-founder tells you when you are wrong. A chatbot tells you that you are brilliant.
 - **It keeps your data yours.** The teaching that travels is generic methodology. Your specifics never leave your machine. You get smarter-from-the-world intelligence without becoming the product.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindrian_os/cli",
-  "version": "1.15.0-beta.5",
+  "version": "1.15.0-beta.9",
   "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",