@mindrian_os/install 1.13.0-beta.11

package/references/brain/schema.md

@@ -0,0 +1,76 @@
+# Brain Schema Reference
+
+Neo4j Aura graph -- 21K+ nodes, 65K+ relationships. Larry's teaching intelligence.
+
+## How Brain Connects
+
+**CLI:** `brain-client.cjs` calls the Brain HTTP API directly using `MINDRIAN_BRAIN_KEY` from `.env` or `~/.mindrian.env`. No MCP config needed. Functions: `brain.query(cypher)`, `brain.search(text)`, `brain.schema()`.
+
+**Desktop/Cowork:** Add Brain as an MCP server in `claude_desktop_config.json` or Cowork settings. Server name: `mindrian-brain`. Tools become:
+- `mcp__mindrian-brain__brain_query` -- execute Cypher queries
+- `mcp__mindrian-brain__brain_write` -- write operations (admin only)
+- `mcp__mindrian-brain__brain_schema` -- introspect schema
+- `mcp__mindrian-brain__brain_search` -- semantic search via Pinecone
+
+Note: Brain is NOT configured in the plugin's `.mcp.json` (that file is for the local mindrian-os stdio server only). Brain config lives in user settings because it requires a per-user API key.
+
+## Node Types
+
+| Label | ~Count | Key Properties | Purpose |
+|-------|--------|----------------|---------|
+| Framework | 275+ | name, description, category, difficulty | Innovation methodology frameworks |
+| Phase | varies | name, order, description, framework_id | Ordered steps within a framework |
+| Concept | varies | name, description | Abstract innovation concepts |
+| ProblemType | 4 | name (Un-Defined, Ill-Defined, Well-Defined, Wicked) | Problem classification |
+| Book | 59 | title, author, isbn | Analyzed source books |
+| Tool | 59 | name, description, pws_mapping | Innovation tools with PWS mappings |
+| Course | varies | name, code, semester | Curriculum courses |
+| Example | 100+ | project_name, grade, grade_numeric, rubric_scores, feedback_patterns, percentile | Graded student work |
+
+## Relationships
+
+| Relationship | From -> To | Properties | Why It Matters |
+|--------------|-----------|------------|----------------|
+| FEEDS_INTO | Framework -> Framework | confidence, transform_description | Framework chaining rules |
+| TRANSFORMS_OUTPUT_TO | Framework -> Framework | transform_type, description | How output becomes input |
+| CO_OCCURS | Framework -> Framework | frequency, context | Natural framework pairings |
+| ADDRESSES_PROBLEM_TYPE | Framework -> ProblemType | effectiveness | Framework-to-problem mapping |
+| HAS_PHASE | Framework -> Phase | order | Phase progressions |
+| PREREQUISITE | Framework -> Framework | strength | What must come first |
+| APPLIED_IN | Framework -> Example | section, quality_score | Real usage with grades |
+| REFERENCES | Book -> Framework | chapter, relevance | Source material links |
+
+## Grading Calibration Data
+
+Example nodes provide calibration for grading:
+- `rubric_scores`: {vision, problem_definition, feasibility, market, completeness}
+- `grade` / `grade_numeric`: letter (A-F) and numeric (0-100)
+- `feedback_patterns`: common tags (Vision-to-Execution Gap, Framework Vomit, etc.)
+- `percentile`: ranking within cohort
+
+Grading Agent compares user's room state against this distribution for calibrated assessment.
+
+## .mcp.json Template
+
+```json
+{
+  "mcpServers": {
+    "neo4j-brain": {
+      "command": "npx",
+      "args": ["-y", "@neo4j/mcp-neo4j"],
+      "env": {
+        "NEO4J_URI": "{uri}",
+        "NEO4J_USER": "{user}",
+        "NEO4J_PASSWORD": "{password}"
+      }
+    },
+    "pinecone-brain": {
+      "command": "npx",
+      "args": ["-y", "@anthropic/pinecone-mcp"],
+      "env": {
+        "PINECONE_API_KEY": "{key}"
+      }
+    }
+  }
+}
+```

package/references/capability-radar/capabilities-index.md

@@ -0,0 +1,241 @@
+# Claude Capabilities Index for MindrianOS
+
+Last updated: 2026-05-05 (folded in Claude Code 2.1.110 → 2.1.128 via /mos:radar --fetch)
+
+This reference is curated. Run `/mos:radar --fetch` to check the Claude Code changelog for recent additions. Adoption candidates from the most recent fetch are tracked in `.planning/seeds/SEED-003-claude-code-2-1-x-capability-adoption.md`.
+
+---
+
+## models
+
+### Opus 4.7 (xhigh effort tier)
+- **What:** Current top-tier model. `/effort` slider exposes a maximum-reasoning xhigh tier. Auto mode available for Max subscribers.
+- **MindrianOS relevance:** Default `executor_model: "opus"` resolves to 4.7. xhigh tier is overkill for most plans but worth it for the load-bearing migration phases (108, 109). `/effort` slider lets users dial reasoning depth without leaving the conversation.
+- **Status:** available
+- **Since:** Claude Code 2.1.111
+
+### Opus 4.6 (1M Context)
+- **What:** Extended thinking, adaptive reasoning, 128K max output tokens, 1M context window
+- **MindrianOS relevance:** Methodology sessions can go deeper with full reference loading viable. Complex pipelines (Discovery, Thesis) can hold complete Room state in context without compression.
+- **Status:** available
+- **Since:** 2026 Q1
+
+### Sonnet 4.6 (200K/1M)
+- **What:** Balanced speed and intelligence. 200K default context, extendable to 1M.
+- **MindrianOS relevance:** Default model for most users. Context compression and selective reference loading are critical. Room intelligence pipeline must stay within budget.
+- **Status:** available
+- **Since:** 2026 Q1
+
+### Haiku 4.5
+- **What:** Fast, cost-efficient model for background and lightweight tasks
+- **MindrianOS relevance:** Potential for non-blocking Room analysis, quick classification tasks, and high-volume insight filing where speed matters more than depth.
+- **Status:** available
+- **Since:** 2025 Q4
+
+---
+
+## code
+
+### Statusline API
+- **What:** Real-time display of context window usage, model info, and cost data in the CLI status bar
+- **MindrianOS relevance:** Powers the context-monitor bridge. Lets users see how much context their Room and methodology sessions consume in real time.
+- **Status:** available
+- **Since:** 2025 Q4
+
+### Hooks (SessionStart, Stop, PostToolUse)
+- **What:** Event-driven plugin behavior triggered at session lifecycle points and after tool usage
+- **MindrianOS relevance:** Core of the intelligence pipeline. SessionStart loads Room state and proactive findings. PostToolUse triggers passive insight filing. SessionStop handles cleanup.
+- **Status:** available
+- **Since:** 2025 Q3
+
+### Hooks Can Invoke MCP Tools Natively (`type: "mcp_tool"`)
+- **What:** Hook entries can declare `type: "mcp_tool"` and call any registered MCP tool directly, without spawning a Node child process
+- **MindrianOS relevance:** Removes the `lib/core/brain-client.cjs` proxy layer. Eliminates a class of "hook timed out at 2000ms" failures. Direct moat deepening per the MWP mandate. Tracked as adoption candidate A2 in SEED-003.
+- **Status:** available
+- **Since:** Claude Code 2.1.118
+
+### PostToolUse `updatedToolOutput` (Tool-Output Rewrite)
+- **What:** PostToolUse hooks can rewrite a tool's output via `hookSpecificOutput.updatedToolOutput` before the model sees it
+- **MindrianOS relevance:** Structural enforcement of Canon Part 8. Sanitize Brain MCP responses to redact accidental user-data echo at runtime. Closes the gap noted in CANON-PHASE-MAP.md (Part 8 row, "PR gate pending"). Tracked as adoption candidate A3 in SEED-003.
+- **Status:** available
+- **Since:** Claude Code 2.1.121
+
+### `/usage` (replaces `/cost` + `/stats`)
+- **What:** Unified usage/cost reporting command; replaces the two prior commands
+- **MindrianOS relevance:** Statusline `📊` token-budget glyph (Phase 106-02) should track the new command's output schema. Single-line update.
+- **Status:** available
+- **Since:** Claude Code 2.1.118
+
+### `/skills` Type-to-Filter
+- **What:** Skills picker now has a type-to-filter search box
+- **MindrianOS relevance:** Improves discoverability of MindrianOS's 30+ skills. No code change required on the plugin side.
+- **Status:** available
+- **Since:** Claude Code 2.1.121
+
+### `/ultrareview` Built-in
+- **What:** Built-in parallel code-review command (also available as non-interactive `claude ultrareview [target]`)
+- **MindrianOS relevance:** May overlap with MindrianOS's own `/ultrareview`. Verify; if Anthropic's built-in is sufficient, delegate to it and remove the MindrianOS duplication.
+- **Status:** available
+- **Since:** Claude Code 2.1.111
+
+### `${CLAUDE_EFFORT}` in Skills
+- **What:** Skills can read `${CLAUDE_EFFORT}` to calibrate verbosity/depth based on the user's current `/effort` setting
+- **MindrianOS relevance:** Larry skills (larry-personality, larry-extended) can match teaching depth to user effort. xhigh effort = deeper Feynman decomposition; low effort = terse navigation.
+- **Status:** available
+- **Since:** Claude Code 2.1.120
+
+### `claude project purge`
+- **What:** Subcommand to delete all Claude Code state at a path with `--dry-run`, `-y`, `-i`, `--all` flags
+- **MindrianOS relevance:** Clean-slate command for testers when an install goes sideways. Mention in tester onboarding docs.
+- **Status:** available
+- **Since:** Claude Code 2.1.126
+
+### Background Tasks (run_in_background)
+- **What:** Async command execution that runs without blocking the main conversation
+- **MindrianOS relevance:** Potential for non-blocking Room analysis, export generation, and Brain queries that don't interrupt the user's flow.
+- **Status:** available
+- **Since:** 2025 Q4
+
+### Max Output Tokens (128K for Opus)
+- **What:** Opus can generate up to 128K tokens in a single response
+- **MindrianOS relevance:** Export and document generation quality. Full thesis documents, comprehensive Room exports, and detailed grading reports can be generated in one pass.
+- **Status:** available
+- **Since:** 2026 Q1
+
+---
+
+## desktop_cowork
+
+### Cowork
+- **What:** Multi-user persistent agents with shared 00_Context/ state directory
+- **MindrianOS relevance:** Team venture collaboration. Multiple users can work on the same venture with shared Room state, concurrent methodology sessions, and unified Data Room.
+- **Status:** available
+- **Since:** 2025 Q4
+
+### Desktop
+- **What:** Conversational interface with MCP server support and visual interaction
+- **MindrianOS relevance:** Larry personality shines here. Users interact naturally without commands. MCP Brain connection works seamlessly. One of the three surfaces MindrianOS must support.
+- **Status:** available
+- **Since:** 2025 Q3
+
+### Agent Teams
+- **What:** Experimental multi-agent coordination for complex tasks
+- **MindrianOS relevance:** Potential for specialized agent collaboration. Research Agent + Investor Agent could run adversarial analysis in parallel. Pipeline stages could delegate to focused agents.
+- **Status:** experimental
+- **Since:** 2026 Q1
+
+### Forked Subagents on External Builds (`CLAUDE_CODE_FORK_SUBAGENT=1`)
+- **What:** True forked subagents enabled outside Anthropic-internal builds via opt-in env var
+- **MindrianOS relevance:** Substrate for Canon Part 2 Engine 2 (BONO Orchestration) — spawning hat-instantiated team members in parallel. Tracked as adoption candidate A4 in SEED-003.
+- **Status:** available (opt-in)
+- **Since:** Claude Code 2.1.117
+
+### Agent Frontmatter `mcpServers` Declaration
+- **What:** Individual agent files can declare required MCP servers in frontmatter; loaded for main-thread sessions via `--agent`
+- **MindrianOS relevance:** Per-agent Brain MCP scoping. `mos-research` can require Brain without polluting global config; `mos-investor` (synthesis hat) can opt out. Tracked in SEED-003 A4.
+- **Status:** available
+- **Since:** Claude Code 2.1.117
+
+### Push Notification Tool for Remote Control
+- **What:** Built-in tool for asynchronously notifying the user (push notification surface)
+- **MindrianOS relevance:** Larry can ping the user when Brain finishes an enrichment cycle, when a proactive scan finds a contradiction, or when a long-running methodology completes. Today these only surface if the user is actively in the session.
+- **Status:** available
+- **Since:** Claude Code 2.1.110
+
+### `/focus` View
+- **What:** Toggleable focus view in the TUI
+- **MindrianOS relevance:** Aligns with Canon's focus-node concept (Phase 109 D-01). Statusline `🎯` glyph should coordinate with `/focus` state — when user is in focus view, render the focus node prominently.
+- **Status:** available
+- **Since:** Claude Code 2.1.110
+
+---
+
+## plugins_mcp
+
+### Plugin Marketplace
+- **What:** Git-based plugin distribution with auto-updates and version management
+- **MindrianOS relevance:** Primary distribution channel. Users install MindrianOS with one command. Auto-updates keep methodology content current.
+- **Status:** available
+- **Since:** 2025 Q4
+
+### `--plugin-dir` Accepts `.zip` Archives
+- **What:** `claude --plugin-dir <path>` now accepts a `.zip` archive in addition to a directory
+- **MindrianOS relevance:** Beta-tester distribution side-channel. Today: marketplace tag is the only sanctioned path; beta gating requires marketplace.json to advertise the version. Zip channel decouples — hand a tester a single `.zip` without touching marketplace state. Tracked as adoption candidate A5 in SEED-003.
+- **Status:** available
+- **Since:** Claude Code 2.1.128
+
+### `claude plugin validate` (Expanded Manifest)
+- **What:** Plugin validation command accepts a wider set of manifest fields
+- **MindrianOS relevance:** `plugin.json` can declare more metadata that the marketplace honors. Audit current manifest against new accepted fields when planning v1.13.0.
+- **Status:** available
+- **Since:** Claude Code 2.1.120
+
+### Plugin Dependency Auto-Resolution
+- **What:** Plugin dependencies are auto-resolved from configured marketplaces
+- **MindrianOS relevance:** Simplifies coupling between MindrianOS and Brain MCP if Brain ever ships as a separate marketplace plugin (currently bundled).
+- **Status:** available
+- **Since:** Claude Code 2.1.117
+
+### MCP Tool Search
+- **What:** Auto-discovers available MCP tools at 10% context threshold
+- **MindrianOS relevance:** Brain MCP discovery. When Brain is connected, Claude automatically finds Neo4j query tools without explicit configuration.
+- **Status:** available
+- **Since:** 2025 Q4
+
+### MCP `alwaysLoad` Server Config
+- **What:** `alwaysLoad: true` in `.mcp.json` skips tool-search deferral and surfaces server tools from turn 1
+- **MindrianOS relevance:** **Eliminates Brain MCP cold-start window.** Today Larry's first session response is "Brain-blind" until the 10% discovery threshold fires. With `alwaysLoad`, Mode A (Full Loop) per Canon Part 3 starts at turn 1. Single biggest leverage adoption per SEED-003 A1.
+- **Status:** available
+- **Since:** Claude Code 2.1.121
+
+### MCP Auto-Retry on Transient Startup Errors
+- **What:** MCP servers auto-retry transient startup errors up to 3 times
+- **MindrianOS relevance:** Brain MCP reliability improvement. Reduces "Brain unavailable" Tier 0 fallbacks caused by network blips during boot.
+- **Status:** available
+- **Since:** Claude Code 2.1.121
+
+### Concurrent MCP Server Connections (Default)
+- **What:** Multiple MCP servers connect concurrently at startup instead of serially
+- **MindrianOS relevance:** Faster session start when MindrianOS users have Brain + other MCP servers configured (Notion, Gmail, etc.). Net positive for plugin UX.
+- **Status:** available
+- **Since:** Claude Code 2.1.117
+
+### `/mcp` Tool Count + Zero-Tool Flagging
+- **What:** `/mcp` command shows the tool count for connected servers and flags servers that connected with 0 tools
+- **MindrianOS relevance:** Direct diagnostic for Brain MCP failure modes. When a tester says "Brain isn't responding," step 1 becomes `/mcp` — if Brain shows 0 tools, the failure mode is now visible. Candidate for `/mos:doctor --brain` integration.
+- **Status:** available
+- **Since:** Claude Code 2.1.128
+
+### Plugin Hooks
+- **What:** SessionStart, SessionStop, and PostToolUse events available to plugins
+- **MindrianOS relevance:** Intelligence pipeline foundation. Every hook fires plugin scripts that maintain Room state, file insights, and surface proactive findings.
+- **Status:** available
+- **Since:** 2025 Q3
+
+### Statusline in Plugins
+- **What:** Plugins can declare custom statusline scripts for persistent display
+- **MindrianOS relevance:** Context-monitor displays real-time Room health and context budget consumption directly in the CLI status bar.
+- **Status:** available
+- **Since:** 2025 Q4
+
+---
+
+## visualization
+
+### Mermaid Rendering
+- **What:** Built-in Mermaid diagram support for flowcharts, graphs, and sequences
+- **MindrianOS relevance:** Data Room relationship visualization. Knowledge graphs, pipeline flows, and venture stage maps rendered inline.
+- **Status:** available
+- **Since:** 2025 Q3
+
+### SVG Generation
+- **What:** Direct SVG output for vector graphics
+- **MindrianOS relevance:** De Stijl visual assets. Room dashboard graphics, profile documents, and export visuals use SVG for clean rendering.
+- **Status:** available
+- **Since:** 2025 Q3
+
+### HTML Artifact Rendering
+- **What:** Browser-based rich output via localhost serving
+- **MindrianOS relevance:** Data Room dashboard. The De Stijl-styled interactive viewer with knowledge graph, chat panel, and methodology output display.
+- **Status:** available
+- **Since:** 2025 Q4

package/references/capability-radar/changelog-cache.md

@@ -0,0 +1,81 @@
+# Changelog Cache
+
+Last fetched: 2026-05-05
+Source: https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md
+
+## Recent Changes Relevant to MindrianOS
+
+### 2.1.128
+- **Domain:** plugins_mcp
+- **Change:** Bare `/color` picks random session color; `/mcp` shows tool count for connected servers and flags servers with 0 tools; `--plugin-dir` accepts `.zip` archives; updated `/model` picker with collapsed Opus 4.7 entries; MCP reconnects summarize re-announced tools by server prefix; `/plugin update` fixed for npm-sourced plugins
+- **MindrianOS impact:** `.zip` distribution is a side-channel for beta testers (Lawrence + 4 new testers in `docs/testers/`) ahead of marketplace tag promotion. `/mcp` 0-tool flagging is direct diagnostic for Brain MCP failure modes — candidate for `/mos:doctor --brain` integration. `/plugin update` npm fix matters if MindrianOS ever ships an npm-sourced variant.
+
+### 2.1.126
+- **Domain:** code
+- **Change:** `claude project purge [path]` deletes Claude Code state with `--dry-run`, `-y`, `-i`, `--all` flags; `--dangerously-skip-permissions` bypasses prompts for protected paths; OAuth code can be pasted when browser callback unreachable; PowerShell improvements on Windows
+- **MindrianOS impact:** `claude project purge` is a clean-slate command for testers when a room install goes sideways. Could replace ad-hoc `~/.claude/plugins/mindrian-os.stale-*` cleanup. PowerShell parity matters if MindrianOS expands to Windows testers.
+
+### 2.1.122
+- **Domain:** plugins_mcp + models
+- **Change:** Added `ANTHROPIC_BEDROCK_SERVICE_TIER` env variable; `/resume` finds session from pasted PR URL; `/mcp` shows hidden claude.ai connectors with deduplication hints
+- **MindrianOS impact:** Bedrock tier selection is relevant for enterprise customers down the road. `/mcp` connector dedup helps when users have both claude.ai-managed MCPs and locally-configured Brain MCP.
+
+### 2.1.121
+- **Domain:** plugins_mcp + code
+- **Change:** Added `alwaysLoad` MCP server config option to skip tool-search deferral; `/skills` now has type-to-filter search box; PostToolUse hooks can replace tool output via `hookSpecificOutput.updatedToolOutput`; MCP servers auto-retry transient startup errors up to 3 times
+- **MindrianOS impact:** **alwaysLoad is the big one for Brain.** Today Brain tools are discovered at the 10% context threshold; `alwaysLoad: true` would surface them from turn 1 — eliminates the "Larry doesn't know about Brain yet" cold-start window. Direct candidate for `.mcp.json` update. PostToolUse `updatedToolOutput` lets hooks rewrite tool output — could enforce Part 8 boundary on Brain query responses (sanitize before they reach the model). MCP auto-retry reduces flake-driven Brain timeouts.
+
+### 2.1.120
+- **Domain:** code + plugins_mcp
+- **Change:** Git for Windows no longer required (PowerShell fallback); `claude ultrareview [target]` non-interactive subcommand; Skills can reference `${CLAUDE_EFFORT}`; `claude plugin validate` accepts more manifest fields
+- **MindrianOS impact:** `${CLAUDE_EFFORT}` in skills lets MindrianOS skills calibrate verbosity (Larry-extended vs quick-larry mode). Plugin validate expansion means `plugin.json` can declare more metadata that the marketplace honors.
+
+### 2.1.119
+- **Domain:** code
+- **Change:** `/config` settings persist to `~/.claude/settings.json`; added `prUrlTemplate` setting; `CLAUDE_CODE_HIDE_CWD` env variable; `--from-pr` accepts GitLab and Bitbucket PRs; `--print` mode honors agent `tools:` and `disallowedTools:` frontmatter
+- **MindrianOS impact:** `--print` mode honoring agent frontmatter means MindrianOS agents (larry-extended, mos-research, etc.) can be invoked non-interactively with their tool restrictions intact — opens scripting/automation paths.
+
+### 2.1.118
+- **Domain:** code + plugins_mcp
+- **Change:** Vim visual and visual-line modes; merged `/cost` and `/stats` into `/usage`; custom themes via `/theme` or `~/.claude/themes/` JSON files; Hooks can invoke MCP tools via `type: "mcp_tool"`; added `DISABLE_UPDATES` env var
+- **MindrianOS impact:** **Hooks → MCP tools is a major shift.** SessionStart, PostToolUse, etc. can now call Brain MCP directly without spawning a child Node process. Could collapse the `lib/core/brain-client.cjs` proxy layer. Custom themes opens De Stijl as a sanctioned theme distribution. `/usage` replacement means token-budget glyph in statusline should track the new command output schema.
+
+### 2.1.117
+- **Domain:** desktop_cowork + plugins_mcp
+- **Change:** Forked subagents enabled on external builds with `CLAUDE_CODE_FORK_SUBAGENT=1`; Agent frontmatter `mcpServers` now loaded for main-thread sessions via `--agent`; `/resume` offers summarization for stale large sessions; concurrent MCP server connections default for faster startup; plugin dependencies auto-resolved from configured marketplaces
+- **MindrianOS impact:** **Forked subagents on external builds** is the unlock for parallel methodology pipelines outside Anthropic infrastructure. Agent-level `mcpServers` declaration means individual agents (e.g. `mos-research`) can require Brain MCP without polluting global config. Plugin deps auto-resolution simplifies MindrianOS Brain MCP coupling.
+
+### 2.1.116
+- **Domain:** plugins_mcp
+- **Change:** `/resume` on large sessions 67% faster with improved fork handling; faster MCP startup; `resources/templates/list` deferred to first `@`-mention; smoother fullscreen scrolling; `/terminal-setup` configures editor scroll sensitivity; thinking spinner shows inline progress
+- **MindrianOS impact:** Faster MCP startup directly improves Brain connection time. The `resources/templates/list` deferral is something to validate against `.mcp.json` — Brain may need to opt into eager template loading if it relies on templates being available at startup.
+
+### 2.1.111
+- **Domain:** models + code
+- **Change:** Claude Opus 4.7 xhigh available via `/effort`; Auto mode for Max subscribers with Opus 4.7; `/effort` opens interactive slider without arguments; "Auto (match terminal)" theme; added `/less-permission-prompts` skill; added `/ultrareview` for parallel code review
+- **MindrianOS impact:** **Opus 4.7 is the current top-tier model.** Default `executor_model: "opus"` in MindrianOS config now resolves to 4.7. The xhigh effort tier is overkill for most plans but worth knowing for the load-bearing migration phases (108, 109). `/ultrareview` is a built-in pattern that could displace any custom MindrianOS multi-reviewer flows.
+
+### 2.1.110
+- **Domain:** code + desktop_cowork
+- **Change:** Added `/tui fullscreen` command; added push notification tool for Remote Control; changed `Ctrl+O` to toggle normal/verbose transcript; `/focus` toggles focus view; added `autoScrollEnabled` config; `--resume`/`--continue` resurrects unexpired scheduled tasks
+- **MindrianOS impact:** Push notification tool opens async user signaling — Larry could ping the user when Brain finishes an enrichment cycle or when a proactive scan finds a contradiction. `/focus` view aligns with the canon's focus-node concept (Phase 109 D-01) — consider statusline coordination so the `🎯` glyph respects `/focus` state.
+
+## New capabilities NOT in capabilities-index.md (last updated 2026-03-22)
+
+The following are recent additions worth merging into the curated index:
+
+1. **`alwaysLoad` MCP config** (2.1.121) — eliminates Brain MCP cold-start window
+2. **Hooks → MCP tools** (2.1.118) — hook scripts can call Brain directly, no Node child process
+3. **Forked subagents on external builds** (2.1.117) — parallel methodology pipelines unlocked
+4. **Agent frontmatter `mcpServers`** (2.1.117) — per-agent Brain MCP scoping
+5. **PostToolUse `updatedToolOutput`** (2.1.121) — Part 8 sanitization layer for Brain responses
+6. **`--plugin-dir` accepts `.zip`** (2.1.128) — beta-tester distribution side-channel
+7. **`/mcp` 0-tool flagging** (2.1.128) — Brain MCP failure-mode diagnostic
+8. **Opus 4.7 + xhigh effort** (2.1.111) — current top model + max-effort tier
+9. **`/ultrareview` built-in** (2.1.111) — built-in parallel code review pattern
+10. **`/effort` slider** (2.1.111) — runtime effort calibration for skills (`${CLAUDE_EFFORT}`)
+11. **`/usage` (replaces `/cost` + `/stats`)** (2.1.118) — statusline token-budget glyph should track new schema
+12. **Custom themes via JSON** (2.1.118) — De Stijl as sanctioned theme distribution
+13. **`claude project purge`** (2.1.126) — clean-slate command for tester recovery
+14. **MCP auto-retry on transient errors** (2.1.121) — Brain reliability improvement
+15. **Concurrent MCP startup** (2.1.117) — faster session start

package/references/causal/causal-schema.md

@@ -0,0 +1,103 @@
+---
+title: Causal Layer Schema Reference
+description: >
+  KuzuDB node and edge schema for the causal reasoning layer.
+  Use this document as context when generating Cypher queries for causal analysis.
+version: 1.7.0
+---
+
+# Causal Layer Schema Reference
+
+> Extension to LazyGraph schema for causal reasoning.
+> CausalClaim nodes and causal edges store extracted causal intelligence.
+
+## Node Types
+
+### CausalClaim
+
+Represents a single cause-effect assertion extracted from a room artifact.
+
+| Property | Type | Default | Description |
+|----------|------|---------|-------------|
+| id | STRING (PK) | -- | Unique claim ID, e.g. causal-0001 |
+| cause | STRING | -- | What produces the effect (max 200 chars) |
+| mechanism | STRING | -- | HOW the cause produces the effect (max 300 chars) |
+| effect | STRING | -- | What happens as a result (max 200 chars) |
+| confidence | DOUBLE | 0.5 | Dynamic: observed=0.7, asserted=0.5, inferred=0.3 initial |
+| evidence | STRING | '[]' | JSON array of supporting artifact IDs |
+| source_artifact | STRING | '' | Primary artifact ID this was extracted from |
+| domain | STRING | 'general' | materials, business, competitive, financial, team, legal, general |
+| falsifiable_prediction | STRING | '' | Testable prediction that would disprove this claim |
+| novelty_score | DOUBLE | 0.0 | Graph neighborhood uniqueness (0-1) |
+| extraction_method | STRING | 'inferred' | observed, asserted, inferred |
+| created | STRING | '' | Date extracted (YYYY-MM-DD) |
+
+## Edge Types
+
+### CAUSES (Artifact -> Artifact)
+Direct causal relationship between artifacts.
+
+| Property | Type | Default |
+|----------|------|---------|
+| mechanism | STRING | '' |
+| confidence | DOUBLE | 0.0 |
+| framework | STRING | '' |
+| direction | STRING | 'forward' |
+
+### ROOT_CAUSE_OF (Artifact -> Artifact)
+Multi-hop root cause chain.
+
+| Property | Type | Default |
+|----------|------|---------|
+| chain_length | INT64 | 1 |
+| intermediate_causes | STRING | '[]' |
+| confidence | DOUBLE | 0.0 |
+| discovery_source | STRING | 'manual' |
+
+### CASCADES_TO (CausalClaim -> CausalClaim)
+Assumption failure propagation between claims.
+
+| Property | Type | Default |
+|----------|------|---------|
+| cascade_type | STRING | 'invalidation' |
+| severity | STRING | 'medium' |
+| path_length | INT64 | 1 |
+
+### EXTRACTED_FROM (CausalClaim -> Artifact)
+Provenance link from claim to source artifact. Every CausalClaim MUST have at least one.
+
+No additional properties.
+
+## Query Patterns
+
+### Chain Traversal (bounded, no ACYCLIC)
+```cypher
+MATCH path = (start:CausalClaim)-[:CASCADES_TO*1..6]->(end:CausalClaim)
+WHERE start.id = $claim_id
+RETURN [n IN nodes(path) | n.id] AS chain,
+       [n IN nodes(path) | n.cause] AS causes,
+       length(path) AS depth
+```
+
+### Cross-Reference: Causal + HSI
+```cypher
+MATCH (c1:CausalClaim)-[:EXTRACTED_FROM]->(a1:Artifact)
+      -[h:HSI_CONNECTION]->(a2:Artifact)<-[:EXTRACTED_FROM]-(c2:CausalClaim)
+WHERE c1.id <> c2.id
+RETURN c1.id, c2.id, h.hsi_score, h.surprise_type
+```
+
+### Cross-Reference: Causal + Reverse Salient
+```cypher
+MATCH (c:CausalClaim)-[:EXTRACTED_FROM]->(a:Artifact)-[:BELONGS_TO]->(s1:Section)
+      -[rs:REVERSE_SALIENT]->(s2:Section)
+RETURN c.id, s1.name, s2.name, rs.innovation_thesis
+```
+
+### Cross-Reference: Causal + Analogy
+```cypher
+MATCH (c1:CausalClaim)-[:EXTRACTED_FROM]->(a1:Artifact)
+      -[an:ANALOGOUS_TO]->(a2:Artifact)<-[:EXTRACTED_FROM]-(c2:CausalClaim)
+WHERE c1.id <> c2.id
+RETURN c1.id, c2.id, an.analogy_distance, an.structural_fitness
+```

package/references/design/email-template-standard.md

@@ -0,0 +1,155 @@
+# MindrianOS Official Email Template Standard
+
+## Brand Name Rules (MANDATORY)
+
+- **Correct:** MindrianOS (capital M, lowercase indrian, capital OS, NO hyphen)
+- **Wrong:** MindianOS, Mindrian-OS, MINDRIANOS, mindrianos, Mindrian OS, MindRianOS
+- **In headers/banners:** MINDRIANOS (all caps) is acceptable ONLY in De Stijl header blocks
+- **In body text:** Always MindrianOS
+
+## Layout Rules (MANDATORY)
+
+### Direction and Alignment
+- `dir="ltr"` on the outer wrapper table
+- `text-align:left` on EVERY td, p, and content element
+- No centered body text. Only center: footer Mondrian bar, footer credits, CTA buttons
+- Fixed width: 600px outer table
+
+### Structure
+```
+1. Mondrian Header (red block + blue block + accent bars)
+2. Personal Message (left-aligned, from a person)
+3. Content Sections (left-aligned, with accent borders)
+4. Install Steps (if applicable, numbered with blue squares)
+5. CTA Button (blue block, centered text only)
+6. Footer (Mondrian mini-bar, credits, centered)
+```
+
+### De Stijl Color Palette
+| Color | Hex | Usage |
+|-------|-----|-------|
+| Background | #0a0a0f | Body, outer wrapper |
+| Surface | #12121a | Code blocks, cards, sections |
+| Border | #1e1e2e | Dividers, separators |
+| Cream | #f5f0e8 | Primary text, headings |
+| Muted | #a09a90 | Secondary text, descriptions |
+| Dim | #666666 | Tertiary text, fine print |
+| Red | #a63d2f | Header block, warnings, critiques |
+| Blue | #1e3a6e | Header block, step numbers, links accent |
+| Yellow | #c4a43c | Accent borders, warnings, API keys |
+| Green | #2d6b4a | Success, shipped features, confirmations |
+| Link Blue | #60a5fa | Hyperlinks, code references |
+| Code Green | #22c55e | Terminal commands in code blocks |
+| Amethyst | #6366f1 | Optional accent (updates, versions) |
+
+### Typography (email-safe)
+| Element | Font | Size | Color | Style |
+|---------|------|------|-------|-------|
+| Header brand | Trebuchet MS, Helvetica, sans-serif | 28-32px | #f5f0e8 | 900 weight, 4px letter-spacing, uppercase |
+| Section heading | Trebuchet MS, Helvetica, sans-serif | 18-20px | #f5f0e8 | 900 weight, 3px letter-spacing, uppercase |
+| Sub-heading | Trebuchet MS, Helvetica, sans-serif | 12-14px | #a09a90 | 900 weight, 2px letter-spacing, uppercase |
+| Body text | Arial, Helvetica, sans-serif | 13-15px | #a09a90 | Normal, 1.7-1.8 line-height |
+| Emphasis text | Arial, Helvetica, sans-serif | 13-15px | #f5f0e8 | Bold |
+| Code inline | Courier New, monospace | 12-14px | #60a5fa | On #12121a background |
+| Code block | Courier New, monospace | 13-15px | #22c55e | On #12121a background, 10-14px padding |
+| Fine print | Arial, Helvetica, sans-serif | 10-11px | #444444 or #666666 | Normal |
+
+### Component Patterns
+
+#### Mondrian Header
+```html
+<table width="600"><tr>
+  <td width="400" style="background:#a63d2f;padding:24px 28px;">
+    <p style="...font-size:28px;...uppercase;color:#f5f0e8;">MINDRIANOS</p>
+  </td>
+  <td width="200" style="background:#1e3a6e;padding:24px 20px;">
+    <p style="...font-size:10px;...color:#a09a90;">VERSION / CONTEXT</p>
+  </td>
+</tr></table>
+<table width="600"><tr>
+  <td width="400" style="background:#c4a43c;height:5px;">&nbsp;</td>
+  <td width="200" style="background:#2d6b4a;height:5px;">&nbsp;</td>
+</tr></table>
+```
+
+#### Accent Border Card (left border)
+```html
+<table width="100%"><tr>
+  <td width="3" style="background:#a63d2f;">&nbsp;</td>
+  <td style="background:#12121a;padding:16px 20px;text-align:left;">
+    CONTENT
+  </td>
+</tr></table>
+```
+Colors: #a63d2f (red/warning), #1e3a6e (blue/info), #c4a43c (yellow/highlight), #2d6b4a (green/success)
+
+#### Step Number
+```html
+<div style="width:28px;height:28px;background:#1e3a6e;text-align:center;line-height:28px;
+  font-family:'Trebuchet MS',sans-serif;font-size:14px;font-weight:900;color:#f5f0e8;">1</div>
+```
+
+#### Code Block
+```html
+<table width="100%"><tr>
+  <td style="background:#12121a;padding:10px 14px;text-align:left;">
+    <code style="font-family:'Courier New',monospace;font-size:13px;color:#22c55e;">command here</code>
+  </td>
+</tr></table>
+```
+
+#### API Key / Highlight Box
+```html
+<table width="100%"><tr>
+  <td width="3" style="background:#c4a43c;">&nbsp;</td>
+  <td style="background:#12121a;padding:14px 16px;text-align:left;">
+    <code style="font-family:'Courier New',monospace;font-size:15px;color:#60a5fa;">key-here</code>
+  </td>
+</tr></table>
+```
+
+#### CTA Button
+```html
+<table width="100%"><tr>
+  <td style="background:#1e3a6e;padding:14px 20px;text-align:center;">
+    <a href="URL" style="font-family:'Trebuchet MS',sans-serif;font-size:14px;font-weight:900;
+      letter-spacing:2px;text-transform:uppercase;color:#f5f0e8;text-decoration:none;">LABEL</a>
+  </td>
+</tr></table>
+```
+
+#### Bullet List (Mondrian colored squares)
+```html
+<table cellpadding="0" cellspacing="0"><tr>
+  <td width="6" style="background:#a63d2f;">&nbsp;</td>
+  <td style="padding:0 0 0 10px;font-family:Arial,sans-serif;font-size:12px;color:#a09a90;text-align:left;">
+    Content here
+  </td>
+</tr></table>
+```
+Rotate colors: #a63d2f, #1e3a6e, #c4a43c, #2d6b4a
+
+#### Footer
+```html
+<table cellpadding="0" cellspacing="0" align="center"><tr>
+  <td width="36" style="background:#a63d2f;height:3px;">&nbsp;</td>
+  <td width="24" style="background:#1e3a6e;height:3px;">&nbsp;</td>
+  <td width="16" style="background:#c4a43c;height:3px;">&nbsp;</td>
+  <td width="16" style="background:#2d6b4a;height:3px;">&nbsp;</td>
+</tr></table>
+<p style="...font-size:11px;color:#a09a90;">Sender Name</p>
+<p style="...font-size:10px;color:#444;">MindrianOS -- PWS Methodology, Prof. Lawrence Aronhime, Johns Hopkins</p>
+```
+
+## Rules
+
+1. **NO border-radius anywhere.** Hard rectangles only. De Stijl.
+2. **NO emoji.** Ever.
+3. **NO rounded buttons.** Square CTA blocks.
+4. **NO background images.** Solid colors only.
+5. **NO web fonts.** Email-safe only (Trebuchet MS, Arial, Courier New).
+6. **ALL text left-aligned** except footer credits and CTA button text.
+7. **ALL tables use inline styles.** No external CSS (email clients strip it).
+8. **600px max width.** Fixed, not responsive (email clients handle poorly).
+9. **Padding: 28px sides.** Consistent throughout.
+10. **Sender is always a person** (Jonathan Sagir, Larry, etc.) -- never "MindrianOS Team."