@mindrian_os/cli 2.0.0-beta.3 → 2.0.0-beta.33

package/skills/memory/SKILL.md

@@ -0,0 +1,212 @@
+---
+name: memory
+description: Inspect and operate on the three memory layers (within-session, across-session, cross-room).
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+compatibility: Requires Claude Code (or a host implementing disable-model-invocation semantics); Tier-1 hook mechanics referenced in this skill.
+help_jtbd: "Inspect what the room remembers about your sessions."
+argument-hint: "[query <jtbd> | cross-room | resume | park <jtbd> | complete <jtbd> | --opt-out]"
+body_shape: E (Action Report)
+hitl_shape: "F.8"
+hitl_why: "The three memory layers are queried and written as an independent set of operations in any order."
+# Phase 267.3-06, ruled in 267.3-CLASSIFICATION.md (Row 23): first delivery at commands/memory.md:68, a cross-layer census of in_flight/parked/completed counts, a state readout rather than an analysis.
+interactive_first_reward: "--none (diagnostic surface)"
+body_shape_detail: 4-zone Shape E for default + park + complete + opt-out; Shape G (Comparison Matrix) for query when Phase 101 ships (Shape E fallback otherwise); Shape G Mode A or Shape E Mode B for cross-room depending on Brain availability; Shape F.6 (or F.1 fallback) for resume picker.
+serves_jtbd: ["audit-room"]
+teaching: "When you need to inspect what the room remembers, /mos:memory shows the three layers: within-session, across-session, and cross-room. Memory locality is a Canon Part 9 invariant."
+locks_operator: null
+min_tier: 0
+concurrency: sequential
+streams_events: false
+disable-model-invocation: false
+canon_parts: [4, 8]
+phase: 103
+ui_reference: skills/ui-system/SKILL.md
+allowed-tools: Bash Read AskUserQuestion
+# --- Phase 143.3 connector frontmatter ---
+connector:
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: context_block
+  sub_mode: memory-inspect
+  posture: hold
+  hierarchy_rank: 10
+  filing: none
+  plan_gated: false
+  web_scope: null
+  surface: F.1
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:memory
+
+Inspect or operate on the three Mindrian memory layers per Canon Part 4 (every choice is graph data) and Canon Part 8 (graph boundary). The three layers are the navigator's working memory: within-session (per-room, this conversation), across-session (per-USER aggregate, survives Claude Code restarts), and cross-room (Brain-mediated when reachable, local synthesis when not).
+
+## Layers
+
+- **Within-session** (Phase 100): `<roomDir>/.mindrian/jtbd-state.json`. Per-room, per-session. Holds the active JTBD + recent transitions + classifier evidence. Owned by `lib/hmi/jtbd-state.cjs`. This command READS it but never writes to it.
+- **Across-session** (Phase 103-02): `~/MindrianRooms/.memory/jtbd-history.json`. Global per-USER aggregation. Per-room arrays of in_flight / parked / completed JTBDs. Survives Claude Code restarts. Backs the "What was I working on last week?" query.
+- **Cross-room** (Phase 103-03): Brain (Mode A) when reachable, plus filesystem scan (Mode B) at all times. Brain holds generic JTBD-graph structure only -- Canon Part 8 boundary preserved. The local synthesis works without Brain.
+
+## Subcommands
+
+The command runs `node $CLAUDE_PLUGIN_ROOT/scripts/memory-command.cjs <subcommand> [args...]` and prints the structured output verbatim. The script is the source of truth; do not paraphrase its output.
+
+| Subcommand | Body shape | What it does |
+|------------|-----------|---------------|
+| `/mos:memory` | Shape E | Overview: current within-session JTBD + counts of in_flight / parked / completed across all rooms. |
+| `/mos:memory query <jtbd>` | Shape G (or E fallback) | Per-JTBD across-room view. Rows = rooms, columns = state. Falls back to Shape E with note when Phase 101 not yet shipped. |
+| `/mos:memory cross-room` | Shape G Mode A / E Mode B | Full cross-room landscape. Mode A enriches with Brain pattern hints; Mode B is local synthesis only. |
+| `/mos:memory resume` | Shape F.1 (or F.6) | Interactive picker of in_flight JTBDs across rooms. Phase 101 promotes the F.1 picker to F.6 (JTBD-aware) when shipped. |
+| `/mos:memory park <jtbd>` | Shape E | Move the named JTBD from in_flight to parked. Manual override of the auto-park heuristic (Phase 103 D-07). |
+| `/mos:memory complete <jtbd>` | Shape E | Move the named JTBD to completed. Manual override; usually auto-detected via cascade pipeline (Phase 103 D-08). |
+| `/mos:memory --opt-out` | Shape E | Disable across-session memory globally. Writes `~/MindrianRooms/.memory/.opt-out` sentinel. Re-enable by removing it. |
+
+## Step 1: Parse the user's intent
+
+Look at the invocation:
+
+- `/mos:memory` (no args) -> overview Shape E
+- `/mos:memory query <jtbd>` -> per-JTBD view
+- `/mos:memory cross-room` -> full landscape
+- `/mos:memory resume` -> in_flight picker
+- `/mos:memory park <jtbd>` -> manual park transition
+- `/mos:memory complete <jtbd>` -> manual complete transition
+- `/mos:memory --opt-out` -> disable across-session memory
+
+## Step 2: Execute
+
+Run via Bash:
+
+```bash
+node "${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/scripts/memory-command.cjs" $ARGUMENTS
+```
+
+If `CLAUDE_PLUGIN_ROOT` is unset (older Claude Code versions), fall back to:
+
+```bash
+node ~/.claude/plugins/mindrian-os/scripts/memory-command.cjs $ARGUMENTS
+```
+
+The script does the work:
+
+1. Resolves the active room from `~/MindrianRooms/.rooms/registry.json`.
+2. Honors the global opt-out sentinel at `~/MindrianRooms/.memory/.opt-out` (D-16) and the per-room sentinel at `<roomDir>/.mindrian/.memory-opt-out` (D-17).
+3. Detects Cowork (`<roomDir>/00_Context/` present) and surfaces a Zone 3 warning per RESEARCH §8 Pitfall 8.
+4. Renders a 4-zone Shape E (default), Shape G (query / cross-room Mode A), Shape F.1 / F.6 (resume), or Shape E confirmation (park / complete / opt-out).
+5. When Phase 101 selectors are not yet on disk, gracefully degrades: Shape G falls back to Shape E with note; Shape F.6 falls back to Shape F.1.
+6. Writes (park / complete) go through `lib/hmi/across-session-memory.cjs` atomic O_EXCL pipeline.
+
+## Step 3: Render the output
+
+The script outputs 4-zone shapes per `skills/ui-system/SKILL.md`. Display the script's stdout directly. Do not re-format. Do not strip ANSI color codes. Do not wrap in markdown code fences -- the renderer is the structural surface.
+
+## Privacy contract (Canon Part 8)
+
+Across-session memory is **per-USER**, NOT per-team. In Cowork, each user has their own `~/MindrianRooms/.memory/` scoped to their home directory. Team-shared continuity belongs in STATE.md Decisions section -- not in this layer. The Brain never sees your room contents; cross-room queries carry only generic JTBD identifiers + sha256 hashes + enum scalars (Phase 90 buildBrainQueryContext chokepoint).
+
+To disable across-session memory:
+- Globally: `/mos:memory --opt-out` (or `touch ~/MindrianRooms/.memory/.opt-out`).
+- Per-room: `touch <roomDir>/.mindrian/.memory-opt-out`.
+
+The script honors both sentinels at every write path. Reads continue to work in opt-out mode (so the Zone 3 warning can render); writes return early with an explicit "(opt-out engaged)" note.
+
+## Examples
+
+### Default overview
+
+```
+-- mindrianos -- memory -- overview --
+
+  Within-session: prepare-pitch
+
+  • in_flight  : 2
+  • parked     : 1
+  • completed  : 4
+
+  ▷ /mos:memory cross-room
+  ▷ /mos:memory resume
+  ▷ /mos:memory query <jtbd>
+```
+
+### Cross-room (Mode B, Brain unreachable)
+
+```
+-- mindrianos -- memory -- cross-room --
+
+  ■ prepare-pitch
+     ├─ mindrianos (in_flight, 2026-04-30)
+     ├─ mindrianos-venture (in_flight, 2026-04-29)
+     └─ jhu-pilot (completed, 2026-04-25)
+
+  ⚠ Brain unreachable; cross-room view local-only
+
+  ▷ /mos:memory resume
+  ▷ /mos:memory query <jtbd>
+```
+
+### Cross-room (Mode A, Brain reachable)
+
+```
+-- mindrianos -- memory -- cross-room --
+
+  ■ prepare-pitch
+     ├─ mindrianos (in_flight, 2026-04-30)
+     └─ mindrianos-venture (in_flight, 2026-04-29)
+
+  Patterns:
+    • prepare-pitch → validate-idea (conf 0.71)
+
+  ▷ /mos:memory resume
+  ▷ /mos:memory query <jtbd>
+```
+
+### Park confirmation
+
+```
+-- mindrianos -- memory -- park --
+
+  ✓ prepare-pitch: in_flight → parked
+
+  ▷ /mos:memory
+```
+
+## Voice rules
+
+When Larry surfaces the output conversationally:
+
+- "Two jobs in flight across your rooms. `prepare-pitch` is alive in mindrianos and mindrianos-venture; `find-bottleneck` was parked last week."
+- "Brain is offline; this is the local-only view. The cross-room patterns block lights up when you reconnect."
+- "Want to resume? `/mos:memory resume` lists the picker."
+
+NEVER:
+- Apologize for the memory layer surfacing what it surfaces. The state IS the truth.
+- Add "I ran the command" framing. The renderer is structural; let it speak.
+- Suggest the user park or complete a JTBD they have not asked about. The user decides.
+
+## Cross-references
+
+- **Canon:** `docs/MINDRIAN-CANON.md` Part 4 (every park / complete is a typed graph edge), Part 8 (Brain holds generic methodology only; LOCAL bytes never leave the room).
+- **Phase 100** (`/mos:jtbd`): the within-session layer this command extends.
+- **Phase 101** (selector library): graceful upgrade path -- Shape G + Shape F.6 light up automatically when the selector library lands.
+- **Phase 102** (render-v2): all subcommands flow through the shared 4-zone render contract.
+- **Phase 103-05** (memory hooks): SessionStart resume nudge + Stop hook persistence build on this command's surface.
+
+## Zone 4 (Action Footer)
+
+After presenting results, suggest next actions:
+
+> Want to pick up where you left off? -> /mos:memory resume
+> Want what other rooms learned? -> /mos:memory cross-room

package/skills/memory-cortex-reach/SKILL.md

@@ -0,0 +1,80 @@
+---
+name: memory-cortex-reach
+description: Reach into your memory cortex when a governing thought goes stale or a contradiction lands
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Bring the memory cortex to a Decision Gate when a governing thought goes stale or a contradiction lands."
+body_shape: F.1
+hitl_shape: "F.1"
+hitl_why: "A memory reach resolves to a single next move on the recalled item."
+# Phase 267.3-06, ruled in 267.3-CLASSIFICATION.md (Row 22, navigator-ruled NEEDS-RULING): first delivery at commands/memory-cortex-reach.md:57, a Decision Gate whose actual cortex content is surfaced by the orchestrator only after approval (the show router sub-case).
+interactive_first_reward: "--none (diagnostic surface)"
+body_shape_detail: One reach surfaced at an F.1 Decision Gate; the navigator approves before any cortex read fires
+serves_jtbd: ["navigate"]
+teaching: "When a governing thought you set has gone stale, or a fresh contradiction has just landed against a claim, your memory cortex has something to tell you. /mos:memory-cortex-reach is the navigator-facing surface the orchestrator dispatches to bring that signal to a Decision Gate. It does not re-implement the cortex; it joins the reach spine so the cortex becomes dispatchable."
+ui_reference: skills/ui-system/SKILL.md
+# --- Phase 143.3 connector frontmatter ---
+connector:
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: cross_room
+  sub_mode: memory-cortex-bridge
+  framework: null
+  posture: push_forward
+  hierarchy_rank: 60
+  filing: memory_event_only
+  plan_gated: false
+  web_scope: null
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:memory-cortex-reach
+
+You are Larry. This command is the navigator-facing reach into the memory cortex (Phase 150, MEM-05, D-05). The memory cortex is the projected LOCAL graph of the navigator's own thinking: governing thoughts (the MINTO spine), claims and assumptions, decisions, and the contradiction edges between them. When the cortex signals that it needs attention -- a governing thought has gone stale, or a fresh contradiction has just landed -- this surface brings that signal to a Decision Gate.
+
+This command JOINS the Phase 143.3 connector spine by declaring exactly ONE `connector:` block (connects_to_spine). It is `connects_to_spine` proof that the cortex is dispatchable: the intelligence-orchestrator reads the generated `data/connector-registry.json`, never a hardcoded table, so this surface routes with ZERO edits to the orchestrator. It does NOT re-implement dispatch and it does NOT mint a 7th reach -- it rides the frozen `cross_room` reach (the cortex reach is a cross-room/memory bridge per D-05).
+
+## What fires this reach
+
+The `sensorMemoryCortex` sensor (lib/core/sensors/sensor-memory-cortex.cjs, SENS-08) fires this reach when the projected cortex signals either condition:
+
+- **A stale governing thought.** The navigator set a governing thought (a MINTO spine sentence) and the room has drifted; the thought no longer governs what the room now holds. Pulling the cortex forward re-grounds the navigator in their own stated spine.
+- **A fresh contradiction.** A newly projected contradiction edge sits against a claim the navigator depends on. Surfacing it early is the wicked-problem-management move: a contradiction near a claim is data, not noise (Canon Part 4).
+
+The sensor reads ONLY the projected cortex signals threaded on the LOCAL ctx (a stale-governing-thought freshness flag and a fresh-contradiction count, both produced by the Wave-2 cortex producers in 150-04). It consumes those signals; it does not produce them. It carries only enum/scalar handles, never cortex prose (Canon Part 8).
+
+## What the reach does (orchestrator-routed)
+
+When the navigator APPROVES at the Decision Gate, the orchestrator fires the cross-room memory-cortex bridge: it surfaces the projected cortex neighborhood around the stale thought or the fresh contradiction, read LOCAL via `lib/core/navigation.cjs` (Canon Part 9: SQL is the local mind). The filing hook is `memory_event_only` -- the reach records that the navigator reached into the cortex as a system-bookkeeping `memory_event` (Canon Part 9 audit-node carve-out), and mints NO new truth-claim node. No claim is confirmed by this reach; the cortex read is a navigation event, not a truth promotion.
+
+## Part 8 boundary
+
+The cortex is LOCAL room data -- governing thoughts, claims, contradictions, decisions. None of it enters a Brain query (Canon Part 8). The connector's `framework` is `null` (this reach resolves no Brain framework) and `web_scope` is `null` (the cortex never reaches the web). The sensor reads only enum/scalar ctx signals; the orchestrator reads the cortex LOCAL via the navigation chokepoint. There is no path by which cortex prose reaches the Brain.
+
+## Tri-Polar surface note
+
+- **CLI:** the F.1 dial renders in the terminal; the orchestrator dispatches the reach; the cortex neighborhood files LOCAL via the navigation chokepoint.
+- **Desktop:** Larry surfaces the stale-thought / fresh-contradiction signal conversationally and asks the navigator whether to reach into the cortex.
+- **Cowork:** the cortex reach surfaces the shared room's projected cortex so a team sees the same stale thought or contradiction.
+
+## Error Handling
+
+Follow the 3-line error pattern:
+
+```
+&#10007; [What failed]
+  Why: [specific reason]
+  Fix: [one command to resolve]
+```

package/skills/models/SKILL.md

@@ -0,0 +1,197 @@
+---
+name: models
+description: View or switch model-routing profiles
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Browse and apply 25 methodology models."
+argument-hint: "[list|set|show]"
+body_shape: C (Information Card)
+hitl_shape: "F.1"
+hitl_why: "The routing profile offers a single next-move selection."
+# Phase 267.3-06, ruled in 267.3-CLASSIFICATION.md (Row 24): first delivery at commands/models.md:88, the resolved profile/model routing table, a configuration-state readout of the plugin's own operator settings.
+interactive_first_reward: "--none (diagnostic surface)"
+serves_jtbd: ["compare-options"]
+teaching: "When you want to see or switch which model routes which kind of work, /mos:models opens the routing profile. Different tasks deserve different models."
+ui_reference: skills/ui-system/SKILL.md
+allowed-tools: Read Write Bash AskUserQuestion
+# --- Phase 172-06 CIRS R1 exclude (Canon Part 11) ---
+connector:
+  excluded: true
+  reason: "Utility command. Inspects or selects model profiles on explicit navigator request; an operator configuration surface, not contextually triggered."
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:models
+
+You are Larry. This command manages model routing profiles that control which AI model each agent uses. Profiles determine cost vs. quality tradeoffs across all 8 MindrianOS agents.
+
+## Usage
+
+- `/mos:models` -- show current profile and resolved models for all agents
+- `/mos:models set <profile>` -- switch profile (quality, balanced, budget)
+- `/mos:models override <agent> <model>` -- override a specific agent's model
+- `/mos:models override <agent> clear` -- remove an agent override
+- `/mos:models set-default <profile>` -- set global default for new rooms (writes ~/.mindrian/defaults.json)
+- `/mos:models reset` -- remove room/.config.json, revert to defaults
+
+## Step 1: Resolve Room
+
+Find the active room by walking up from `$PWD` looking for `STATE.md` inside a `room/` directory.
+
+- If `room/STATE.md` exists in the current workspace, the room directory is `room/`.
+- If no room found, show global defaults only (from `~/.mindrian/defaults.json` or hardcoded quality fallback).
+
+## Step 2: Parse Arguments
+
+Determine the subcommand from the user's message:
+
+| Input | Subcommand |
+|-------|-----------|
+| `/mos:models` (no args) | view |
+| `/mos:models set quality` | set |
+| `/mos:models override grading opus` | override |
+| `/mos:models override grading clear` | override-clear |
+| `/mos:models set-default balanced` | set-default |
+| `/mos:models reset` | reset |
+
+## Step 3: Execute Subcommand
+
+### View (default)
+
+Run the model-profiles.cjs table command to get the formatted profile table:
+
+```bash
+node "${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/lib/core/model-profiles.cjs" table <roomDir>
+```
+
+If `CLAUDE_PLUGIN_ROOT` is not set, resolve relative to this command file's directory (one level up from `commands/`).
+
+Display the output with this structure:
+
+```
+╭─ MODEL PROFILES ── Agent Routing ─────────────────────────╮
+│                                                             │
+
+  Room:     {room path or "no room detected"}
+  Profile:  {model_profile from .config.json or "quality (default)"}
+  Stage:    {venture stage from STATE.md or "none detected"}
+
+  {formatted table from model-profiles.cjs}
+
+  Overrides: {list from model_overrides or "none"}
+
+  Tip: /mos:models set balanced    Switch to balanced profile
+       /mos:models override <agent> <model>    Override one agent
+
+│                                                             │
+╰─────────────────────────────────────────────────────────────╯
+```
+
+### Set Profile
+
+1. Validate `<profile>` is one of: quality, balanced, budget
+   - If invalid, show error:
+     ```
+     x Invalid profile: {input}
+       Why: Must be one of: quality, balanced, budget
+       Fix: /mos:models set balanced
+     ```
+
+2. Read room/.config.json (or create empty object if it does not exist)
+3. Set `model_profile` to the new value
+4. Write back with `JSON.stringify(config, null, 2)` via Write tool
+5. Confirm the change:
+   ```
+   [OK] Profile set to {profile}
+
+   Run /mos:models to see resolved models for all agents.
+   ```
+
+### Override Agent
+
+1. Validate `<agent>` is one of the 8 MODEL_PROFILES agent names:
+   larry-extended, framework-runner, grading, investor, brain-query, research, opportunity-scanner, persona-analyst
+
+   If invalid, show error:
+   ```
+   x Unknown agent: {input}
+     Why: Must be one of: larry-extended, framework-runner, grading, investor, brain-query, research, opportunity-scanner, persona-analyst
+     Fix: /mos:models override grading opus
+   ```
+
+2. Validate `<model>` is one of: opus, sonnet, haiku, clear
+   - `clear` removes the override (delete the key from model_overrides)
+   - If invalid, show error:
+     ```
+     x Invalid model: {input}
+       Why: Must be one of: opus, sonnet, haiku (or clear to remove override)
+       Fix: /mos:models override grading sonnet
+     ```
+
+3. Read room/.config.json (or create empty object)
+4. If `<model>` is `clear`:
+   - Delete `config.model_overrides[agent]`
+   - Confirm: `[OK] Override removed for {agent}. Using profile default.`
+5. Else:
+   - Set `config.model_overrides = config.model_overrides || {}`
+   - Set `config.model_overrides[agent] = model`
+   - Confirm: `[OK] {agent} overridden to {model}`
+6. Write back with `JSON.stringify(config, null, 2)`
+
+### Set Global Default
+
+1. Validate `<profile>` is one of: quality, balanced, budget
+   - If invalid, show same error as Set Profile
+
+2. Create the global config directory:
+   ```bash
+   mkdir -p ~/.mindrian/
+   ```
+
+3. Write `{"model_profile":"<profile>"}` to `~/.mindrian/defaults.json` via Write tool
+
+4. Confirm:
+   ```
+   [OK] Global default set to {profile}
+
+   New rooms will use {profile} unless overridden by room/.config.json.
+   ```
+
+### Reset
+
+1. If room/.config.json exists, delete it:
+   ```bash
+   rm room/.config.json
+   ```
+
+2. Confirm:
+   ```
+   [OK] Room config removed. Using quality defaults.
+
+   Profile: quality (hardcoded default)
+   Overrides: none
+   ```
+
+3. If room/.config.json does not exist, tell the user:
+   ```
+   Room already using defaults. No .config.json to remove.
+   ```
+
+## Voice Rules
+
+- Larry's voice. Terse, structural, confident.
+- NO EMOJI. Use only the 12 glyphs from the symbol vocabulary.
+- **Banned phrases (per D-23):** "Great question!", "I'd be happy to help", "It's important to note", "Let me explain", sentences starting with "I", "Here's what I found"
+- Lead with data, not commentary.

package/skills/mos/SKILL.md

@@ -0,0 +1,160 @@
+---
+name: mos
+description: State-aware router that picks the right next surface for the navigator
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Meet the navigator where they are: onboard, status, or next move."
+body_shape: E
+hitl_shape: "F.1"
+hitl_why: "The entry command offers one next move into the system."
+# Phase 267.3-06, ruled in 267.3-CLASSIFICATION.md (Row 25): first delivery at commands/mos.md:97, the routing envelope naming the resolved target; a pure dispatcher over onboard/status/suggest-next with zero net-new behavior of its own (the router sub-case).
+interactive_first_reward: "--none (diagnostic surface)"
+body_shape_detail: Action Report rendering of the routing decision (zone 2 names the resolved target + reason; zone 4 routes to that target as primary)
+serves_jtbd: ["explore"]
+teaching: "When you do not know which /mos:* surface to invoke, /mos:mos picks the right one. No room: onboard. Mostly empty room: status with a next-move hint. Populated room: suggest-next."
+canon_parts: [3, 7]
+phase: 121.5-08
+ui_reference: skills/ui-system/SKILL.md
+allowed-tools: Bash Read AskUserQuestion
+# --- Phase 172-06 CIRS R1 exclude (Canon Part 11) ---
+connector:
+  excluded: true
+  reason: "Utility command. The /mos namespace entry / index surface; a meta dispatcher the navigator invokes directly, carrying no problem-state trigger of its own."
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:mos
+
+You are Larry. /mos:mos is the state-aware router. It picks the right next surface for the navigator based on the current room state.
+
+D-10 LOCKED: a single command that meets the navigator where they are. New navigator (no room) goes to onboard. Mostly-empty room goes to status with a next-move hint. Populated room goes to suggest-next.
+
+## Why /mos:mos exists
+
+Per Cluster 5 audit (2026-05-15): /mos:mos was declared in plugin discovery but the backing file was absent. Per D-10, /mos:mos is the canonical "I do not know which /mos:* to invoke -- pick the right one for me" entrypoint. It is the state-aware router that collapses the navigator's mental load.
+
+Per Canon Part 7 (Reuse Before Build): /mos:mos does NOT add a new behavior. It dispatches to /mos:onboard, /mos:status, or /mos:suggest-next. The router replaces the mental cost; the underlying surfaces are unchanged.
+
+## Algorithm (per D-10)
+
+Step 1: resolve room state.
+
+```bash
+# Read STATE.md if present, otherwise treat as no-room.
+if [ -f STATE.md ]; then
+  ROOM_STATE_EXISTS=true
+  STAGE=$(grep -E '^stage:' STATE.md | head -1 | sed 's/^stage:\s*//')
+  SECTIONS_COUNT=$(ls -d */ 2>/dev/null | grep -v node_modules | wc -l)
+else
+  ROOM_STATE_EXISTS=false
+  STAGE=""
+  SECTIONS_COUNT=0
+fi
+```
+
+Then call the pure-function router:
+
+```bash
+node "${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/lib/core/state-aware-router.cjs" <<EOF
+# Not invokable as a CLI; require the module from a Node script or:
+node -e "
+  const r = require('${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/lib/core/state-aware-router.cjs');
+  const out = r.resolveNextSurface({
+    roomState: {
+      exists: ${ROOM_STATE_EXISTS},
+      sectionsCount: ${SECTIONS_COUNT},
+      stage: '${STAGE}',
+    },
+  });
+  console.log(JSON.stringify(out));
+"
+EOF
+```
+
+The router returns `{route, reason, addendum}`:
+
+- `no_room` -> route is `/mos:onboard`. The navigator gets onboarded.
+- `mostly_empty` -> route is `/mos:status`, addendum is `suggest next move`. Show the room state, then suggest the next move.
+- `populated` -> route is `/mos:suggest-next`. Surface the next-move recommendation directly.
+
+Step 2: emit a single Action Report envelope (Shape E per skills/ui-system/SKILL.md Section 1):
+
+```
+-- MindrianOS -- mos -- routing --
+
+  Routing to: <route> (<reason>)
+  <addendum line, if any>
+
+  -> <route>                 # the resolved next surface
+```
+
+The header names the room (or "MindrianOS / no room" when none exists). Zone 2 names the resolved route + the reason from the router. Zone 4 surfaces the route as the primary action.
+
+Step 3: invoke the target command's behavior. Do NOT echo the routing decision and then ask the user to type the target command -- the routing IS the response. Larry continues with the target command's content directly in the same turn.
+
+## Examples
+
+### Case 1: navigator has no room
+
+```
+-- MindrianOS -- mos -- no-room --
+
+  Routing to: /mos:onboard (no_room)
+
+  -> /mos:onboard           # walkthrough + first room
+```
+
+Then Larry continues with /mos:onboard Step 1 in the same turn.
+
+### Case 2: room exists but is mostly empty
+
+```
+-- acme-robotics -- mos -- mostly-empty --
+
+  Routing to: /mos:status (mostly_empty)
+  suggest next move
+
+  -> /mos:status            # current room state
+```
+
+Then Larry continues with /mos:status output + a Shape F.1 Next Move selector (per Canon Part 3) inviting the navigator's next verb.
+
+### Case 3: room populated
+
+```
+-- acme-robotics -- mos -- populated --
+
+  Routing to: /mos:suggest-next (populated)
+
+  -> /mos:suggest-next      # the next-move recommendation
+```
+
+Then Larry continues with /mos:suggest-next output.
+
+## Cross-references
+
+- `lib/core/state-aware-router.cjs` -- the pure-function router (zero side-effects, unit-tested).
+- `commands/onboard.md` -- the no-room branch destination.
+- `commands/status.md` -- the mostly-empty branch destination.
+- `commands/suggest-next.md` -- the populated branch destination.
+- Canon Part 3 -- the Decision Gate that the target commands' F.1 selectors honor.
+- Canon Part 7 -- reuse-before-build: /mos:mos delegates, does not duplicate.
+
+## Voice rules
+
+- Terse, structural, confident. No filler. No em-dashes.
+- The routing decision IS the response. Never ask the user "shall I run X?" -- just run X.
+- Symbol vocabulary: only the 12 approved glyphs from skills/ui-system/SKILL.md.
+- Error pattern (only if the router throws or STATE.md is unreadable): 3 lines -- What / Why: reason / Fix: /mos:command.