feat(memory): add proactive auto-remember triggers and session distill

.agent/AGENTS.md

@@ -95,6 +95,27 @@ Cross-session SQLite FTS5 memory. Commands: `/remember {content}`, `/recall {que
 
 **Full docs**: `memory/README.md`
 
+### MANDATORY: Proactive Memory Triggers
+
+**You MUST suggest `/remember` when you detect these patterns:**
+
+| Trigger | Memory Type | Example |
+|---------|-------------|---------|
+| Solution found after debugging | `WORKING_SOLUTION` | "That fixed it! Want me to remember this?" |
+| User states a preference | `USER_PREFERENCE` | "I'll remember you prefer tabs over spaces" |
+| Workaround discovered | `WORKING_SOLUTION` | "This workaround worked - should I save it?" |
+| Failed approach identified | `FAILED_APPROACH` | "That didn't work - remember to avoid this?" |
+| Architecture decision made | `DECISION` | "Good decision - want me to remember why?" |
+| Tool configuration worked | `TOOL_CONFIG` | "That config worked - save for next time?" |
+
+**Format**: After detecting a trigger, suggest:
+
+```text
+Want me to remember this? /remember {concise description}
+```
+
+**Do NOT wait for user to ask** - proactively offer to remember valuable learnings.
+
 ## Inter-Agent Mailbox
 
 TOON-based async communication between parallel agent sessions.

.agent/scripts/commands/remember.md

@@ -109,21 +109,72 @@ AI: Storing memory:
     1. Confirm  2. Change type  3. Edit  4. Cancel
 ```
 
-## Auto-Remember Triggers
+## Auto-Remember Triggers (MANDATORY)
 
-AI assistants should proactively suggest `/remember` when:
+AI assistants **MUST** proactively suggest `/remember` when detecting these patterns:
 
-1. A solution is found after debugging
-2. User explicitly states a preference
-3. A workaround is discovered for a tool limitation
-4. An architecture decision is made
-5. A failed approach is identified
+### Conversation Pattern Detection
 
+| User Says | Trigger Type | Memory Type |
+|-----------|--------------|-------------|
+| "that fixed it", "it works now", "solved" | Solution found | `WORKING_SOLUTION` |
+| "I prefer", "I like", "always use", "never use" | Preference stated | `USER_PREFERENCE` |
+| "don't do X", "X doesn't work", "avoid X" | Failed approach | `FAILED_APPROACH` |
+| "let's go with", "decided to", "we'll use" | Decision made | `DECISION` |
+| "the trick is", "workaround", "hack" | Workaround found | `WORKING_SOLUTION` |
+| "configure X as", "set X to", "X needs Y" | Tool config | `TOOL_CONFIG` |
+
+### Required Response Format
+
+When a trigger is detected, immediately offer:
+
+```text
+That worked! Want me to remember this for future sessions?
+
+/remember {concise, actionable description}
+
+(Reply 'y' to confirm, or edit the description)
+```
+
+### Examples
+
+**After fixing a bug:**
 ```text
-AI: That fixed it! Want me to remember this solution for future sessions?
-    /remember {suggested content}
+User: Adding --legacy-peer-deps fixed the npm install
+AI: That worked! Want me to remember this for future sessions?
+
+    /remember Use --legacy-peer-deps flag when npm install fails with peer dependency conflicts
+
+    (Reply 'y' to confirm, or edit the description)
 ```
 
+**When user states preference:**
+```text
+User: I always use pnpm instead of npm
+AI: Got it! Want me to remember this preference?
+
+    /remember User prefers pnpm over npm for package management
+
+    (Reply 'y' to confirm)
+```
+
+**After failed approach:**
+```text
+User: That approach with setTimeout didn't work at all
+AI: Noted. Want me to remember to avoid this approach?
+
+    /remember Avoid using setTimeout for async coordination - use proper async/await or Promises instead
+
+    (Reply 'y' to confirm)
+```
+
+### Do NOT Wait
+
+- Do NOT wait for user to explicitly ask to remember
+- Do NOT skip the suggestion if the learning seems minor
+- DO suggest immediately when pattern is detected
+- DO make the suggestion concise and actionable
+
 ## Storage Location
 
 Memories are stored in SQLite with FTS5 for fast search:

.agent/scripts/commands/session-review.md

@@ -45,22 +45,41 @@ Verify aidevops best practices:
 | TODO tracking | Tasks logged appropriately | Recommended |
 | Quality checks | Linters run before commit | Recommended |
 
-### Step 4: Identify Knowledge to Capture
+### Step 4: Auto-Distill Session Learnings
 
-Look for learnings that should be preserved:
+**MANDATORY**: Run session distillation to automatically extract and store learnings:
+
+```bash
+~/.aidevops/agents/scripts/session-distill-helper.sh auto
+```
+
+This will:
+1. Analyze git commits for patterns (fixes, features, refactors)
+2. Extract learnings with appropriate types (ERROR_FIX, WORKING_SOLUTION, etc.)
+3. Store them to memory automatically
+
+### Step 5: Identify Additional Knowledge to Capture
+
+After auto-distill, look for learnings that may have been missed:
 
 1. **Corrections Made**: Did the AI make mistakes that were corrected?
 2. **New Patterns**: Were new approaches discovered?
 3. **Tool Issues**: Did any tools not work as expected?
 4. **User Preferences**: Did the user express preferences?
 
-For each learning, suggest where to document:
+For each additional learning, suggest:
+
+```text
+/remember {concise description of learning}
+```
+
+Or document in:
 - Agent improvements → `@agent-review`
 - Code patterns → Code comments or docs
 - User preferences → `memory/` files
 - Temporary workarounds → TODO.md
 
-### Step 5: Session Health Assessment
+### Step 6: Session Health Assessment
 
 Determine recommendation: