sync Claude Code command with v4 features

.claude/commands/remember.md

@@ -1,18 +1,36 @@
 ---
-description: Commit this conversation to working memory. Extracts knowledge, decisions, insights, open threads, and next actions.
+description: Commit this conversation to working memory. Extracts knowledge, decisions, insights, open threads, and next actions — then writes a living summary and changelog entry for the detected project node. Also flushes any observations accumulated by the passive learning engine. Can run in full mode (user-triggered) or silent mode (auto-triggered at conversation end).
 ---
 
 # /remember
 
-You are committing this conversation to working memory. Work through the following steps precisely.
+You are committing this conversation to Claude's working memory. Work through the following steps precisely.
+
+---
+
+## Mode Detection
+
+This command runs in one of two modes:
+
+### Full mode (default — user triggered /remember explicitly)
+- Complete extraction with all steps below
+- Confirmation output at the end
+- User sees what was committed
+
+### Silent mode (auto-triggered at conversation end)
+- Same extraction logic, but NO confirmation output
+- Triggered when the conversation is ending (farewell signals detected)
+- Skip if the conversation was trivial (no decisions, no knowledge, no meaningful observations)
+- Exception: if creating a NEW node, briefly mention: "Noted — started tracking [node-id]."
+- Always commit user observations (corrections, preferences) even from trivial conversations — these are never wasted
 
 ---
 
 ## Step 1 — Detect the project node
 
 Scan the full conversation and identify which project(s) this session belongs to.
 
-Check existing memory for any established project nodes first. If a matching node exists, use it. If the session covers something genuinely new, create a new node using kebab-case.
+Check Claude's existing memory for any established project nodes first. If a matching node exists, use it. If the session covers something genuinely new, create a new node using kebab-case.
 
 **Node naming conventions:**
 - Client work: prefix with `client:` (`client:acme-corp`, `client:northstar`)
@@ -93,30 +111,36 @@ INSIGHTS:
 - New understanding gained during this session
 - "Aha" moments, reframes, or realizations
 - Connections spotted between previously unrelated things
+- Example: "Realized our churn isn't a product problem — it's an onboarding problem. Customers who complete setup in week 1 retain at 92%."
 
 LESSONS LEARNED:
 - Things tried that worked → why they worked
 - Things tried that failed → why they failed, what to do instead
 - Mistakes made → what the correct approach is
+- Example: "Tried sending proposals same-day — close rate dropped. Waiting 48 hours with a tailored deck closes 3x better."
 
 MENTAL MODELS:
 - How something works, explained for future reference
 - Business processes, decision frameworks, domain rules, workflows
 - "The way X actually works is..."
+- Example: "Their approval process: department head → finance review → VP sign-off → procurement. Skip finance and it stalls for weeks."
 
 GOTCHAS & PITFALLS:
 - Specific traps or non-obvious behaviors in processes, tools, or relationships
 - Things that look like they should work but don't
 - Surprising rules, undocumented requirements, edge cases
+- Example: "Their fiscal year starts in April, not January — all budget conversations need to reference Q1 as April-June."
 
 PATTERNS & RECIPES:
 - Techniques or approaches that proved effective
 - Reusable solutions to specific types of problems
 - Workflows or processes worth repeating
+- Example: "For getting executive buy-in: lead with the metric they own, show the gap, propose one action, name the timeline."
 
 CORRECTED BELIEFS:
 - Assumptions that turned out to be wrong
 - Previous understanding that was updated or reversed
+- Example: "Previously assumed they were price-sensitive — actually they have budget, they just need ROI framing to get internal approval."
 
 CONTEXT & BACKGROUND:
 - Domain knowledge needed to understand this project or topic
@@ -131,6 +155,7 @@ CONTEXT & BACKGROUND:
 CROSS-PROJECT SIGNALS:
 - Anything with implications for a different node
 - Knowledge that applies across projects
+- "The caching pattern here would solve the performance issue in [other-node]"
 
 PEOPLE:
 - Anyone who came up: name, role, context
@@ -143,11 +168,39 @@ QUESTIONS FOR NEXT TIME:
 - Hypotheses to confirm or reject
 ```
 
+### D. User Observations (from passive learning engine)
+
+```
+PREFERENCES:
+- Communication style signals (terse vs. detailed, options vs. decisions)
+- Tool/platform preferences expressed or demonstrated
+- Working style patterns (batching, time-of-day, delegation)
+- Vocabulary and framing they use and expect
+
+CORRECTIONS:
+- Anything the user corrected about your approach
+- Implicit corrections (they ignored your suggestion and did it differently)
+- Format: [what was wrong] → [what they wanted] → [why, if given]
+
+DOMAIN CONTEXT:
+- Facts about their company, team, industry shared in passing
+- How their processes/systems work
+- Constraints they operate under
+```
+
+These observations are ALWAYS extracted, even in silent mode. They go to the `user` node, not project nodes (unless the observation is project-specific domain knowledge).
+
 ---
 
 ## Step 3 — Write to memory
 
-Memory is stored at `~/Documents/Claude/memory/`. Create directories as needed.
+### Storage Location
+
+Memory is stored in `~/Documents/Claude/memory/`.
+
+**Before writing**: Ensure the memory directory exists (`~/Documents/Claude/memory/` on macOS/Linux, or `%USERPROFILE%\Documents\Claude\memory` on Windows). Create parent folders if missing.
+
+If the directory cannot be created or accessed, explain that memory cannot be persisted without this folder and stop.
 
 1. Determine the file path from the node ID:
    - If node has a prefix (e.g., `client:acme-corp`): `memory/{prefix}/{slug}.md`
@@ -162,6 +215,35 @@ Memory is stored at `~/Documents/Claude/memory/`. Create directories as needed.
    - Add any new knowledge entries to Recent Knowledge (keep last 7 days only)
    - Update the "Last updated" timestamp
 
+#### Dashboard File Format
+
+If `DASHBOARD.md` doesn't exist, create it with this template:
+
+```markdown
+# Working World Dashboard
+> Last updated: YYYY-MM-DD
+
+## Active Nodes
+| Node | Summary | Last Updated |
+|------|---------|-------------|
+| [node-id] | [1-line living summary] | YYYY-MM-DD |
+
+## P0 Actions
+- [P0] [node-id]: [action]
+
+## Waiting On
+- [WAITING:who] [node-id]: [what]
+
+## Recent Knowledge (last 7 days)
+- [node-id] [TYPE] (date): [entry]
+
+## Stale Threads
+- [node-id]: [thread] — open since [date]
+
+## Dormant Nodes
+- [node-id]: Last active [date]
+```
+
 #### Node File Format
 
 ```markdown
@@ -210,6 +292,45 @@ Memory is stored at `~/Documents/Claude/memory/`. Create directories as needed.
 [node] SIGNAL from [other-node] (date): [implication]
 ```
 
+### User Node — Observation Flush
+
+**Before writing project data**, flush observations to the user profile node:
+
+1. File path: `~/Documents/Claude/memory/user.md`
+2. If it doesn't exist, create it with this template:
+
+```markdown
+# user
+> Last updated: YYYY-MM-DD
+
+## Communication Preferences
+
+## Working Style
+
+## Corrections
+
+## Domain Expertise
+
+## Relationships & Org Context
+
+## Tool & Platform Preferences
+```
+
+3. Append new observations to the appropriate section:
+   - Preferences → Communication Preferences or Working Style or Tool & Platform Preferences
+   - Corrections → Corrections (always — these are highest priority)
+   - Domain context → Domain Expertise
+   - Relationship info → Relationships & Org Context
+
+4. Before appending, check for duplicates or contradictions:
+   - If new observation reinforces existing → skip (already captured)
+   - If new observation contradicts existing → write a CORRECTION entry, remove the old one
+   - If new observation is genuinely new → append
+
+5. Update `DASHBOARD.md` with the user node's last-updated timestamp
+
+**Confidence gating**: Only commit observations with reasonable confidence. A single offhand remark is a signal, not a fact. But direct corrections — always commit.
+
 ### A. Living Summary (replace if exists, create if new)
 
 Format:
@@ -281,19 +402,29 @@ Scan for:
 
 ## Step 5 — Confirm
 
+### Full mode (user-triggered)
+
 Respond with:
 1. **Node(s)** written to
 2. **Living summary** (for verification)
 3. **Knowledge captured** — list INSIGHT/LESSON/MODEL/GOTCHA/RECIPE/CORRECTION entries
-4. **Open threads** with staleness
-5. **Blockers** (if any)
-6. **Next actions** — P0 → P1 → P2 → WAITING
-7. **Cross-project signals** (if any)
-8. **Knowledge updates** — prior entries invalidated, reinforced, or corrected
-9. **Unanswered questions** carried forward
+4. **Observations captured** — count of preferences, corrections, patterns written to user node
+5. **Open threads** with staleness
+6. **Blockers** (if any)
+7. **Next actions** — P0 → P1 → P2 → WAITING
+8. **Cross-project signals** (if any)
+9. **Knowledge updates** — prior entries invalidated, reinforced, or corrected
+10. **Unanswered questions** carried forward
 
 Keep tight. Skimmable in 20 seconds.
 
+### Silent mode (auto-triggered)
+
+- Say nothing unless creating a new node
+- If creating a new node: "Noted — started tracking [node-id]."
+- If a critical correction was captured that changes future behavior, optionally note: "Got it — I'll [adjusted behavior] going forward."
+- Otherwise: complete silence. The user shouldn't notice the commit happened.
+
 ---
 
 ## Memory hygiene

CHANGELOG.md

@@ -4,6 +4,11 @@ All notable changes to the Cortex Plugin are documented here.
 
 Format follows [Keep a Changelog](https://keepachangelog.com/). Versions match `plugin.json`.
 
+## [Unreleased]
+
+### Fixed
+- `.claude/commands/remember.md` now includes v4 features: silent mode, user observation extraction, user node writes, and dashboard template (parity with `commands/remember.md`).
+
 ## [4.0.0] — Always-On Learning
 
 ### Added