chore: auto-commit uncommitted changes

HEARTBEAT.md

@@ -59,6 +59,39 @@ sessions_spawn(task="inou daily suggestion: review memory/inou-context.md and re
 
 ---
 
+### Cron Job Failure Check (every heartbeat)
+Check if systemd timers/services have failed since last run:
+```bash
+systemctl --user list-units --state=failed --no-legend 2>/dev/null
+```
+Key timers to watch: `daily-updates.timer`, `k2-watchdog.timer`
+
+If any service shows `Result=exit-code` or `ActiveState=failed`:
+1. Check logs: `journalctl --user -u <service> -n 50 --no-pager`
+2. If it's `daily-updates.service` failing: check `memory/updates/YYYY-MM-DD.json` for what ran (nightly maintenance may have covered it), then fix the script
+3. **Alert via ntfy (forge-alerts)** — don't wait for Johan to notice
+
+```bash
+# Quick check
+systemctl --user show daily-updates.service --property=Result,ActiveState | grep -v "^$"
+```
+
+---
+
+### Disk Usage Check (every heartbeat)
+```bash
+df -h / | awk 'NR==2 {print $5, $4}'
+```
+- **< 70%:** Fine, no action
+- **70–85%:** Note in next briefing
+- **> 85%:** Alert via ntfy (forge-alerts) immediately
+- **> 90%:** URGENT — ping Johan
+
+Current baseline: ~54% used (236G / 469G) as of 2026-03-15.
+Top consumers to check if filling fast: ClickHouse (`/var/lib/docker`), Jellyfin media, doc inbox.
+
+---
+
 ### Memory Review (daily)
 Quick scan of today's conversations for durable facts worth remembering:
 - New people, relationships, preferences

drafts/vault1984-pr-plan.md

@@ -35,6 +35,22 @@ Not a blog post. A letter. With the list of laws, the architecture argument, and
 
 ---
 
+## High-value X targets
+
+**Peter Steinberger (@steipete)**
+Founder of OpenClaw, just joined OpenAI to "bring agents to everyone." Actively amplifies tools built around OpenClaw. 5.3M views on his OpenAI announcement tweet. Not a cold pitch — engage when vault1984 is the natural answer to "my OpenClaw agent needs credentials." The connection: vault1984 is the credential layer for exactly what he's building.
+
+**Chao Huang (@huang_chao4969)**
+CLI-Anything — "making ALL software agent-native." 11K GitHub stars in 5 days, 18.2K views on this tweet. The connection: CLI-Anything makes any software agent-controllable. Every agent controlling software needs credentials. vault1984 is the answer to the problem CLI-Anything creates at scale. Engage in the #clianything / #AIAgents threads when vault1984 ships. Natural reply, not a cold pitch.
+
+**Brian Krebs (@briankrebs)**
+krebsonsecurity.com. Most read security journalist. Covered LastPass breach exhaustively. Pitch the LastPass page + architecture when product ships.
+
+**Troy Hunt (@troyhunt)**
+HaveIBeenPwned.com. THE breach authority. One mention reaches every security professional. Architecture argument is his language.
+
+---
+
 ## Distribution plan (when ready)
 
 ### Anchor

drafts/x-inou-promotion-plan.md

@@ -209,3 +209,23 @@ Example workflow:
 ---
 
 *Ready to discuss when you wake up.*
+
+---
+
+## Pitch Angles — Strong Examples Found
+
+### Hans Amato thread (Mar 12, 2026) — 13k likes, 2k RTs
+https://x.com/HansAmato/status/2032139819313480071
+
+The story: $12k therapy for anger → real cause was gut leakage → inflammation → cortisol → low magnesium → no GABA. Fixed with data, not talk.
+"3 practitioners. 2 years. $12,000+. Zero blood draws."
+
+**Why it works for inou:**
+- Visceral, relatable (not a rare tragedy — everyday dysfunction)
+- Demonstrates the fragmentation problem at scale
+- Proves the market gets it — 13k people immediately understood
+- Lower emotional barrier than origin stories involving serious injury
+- The fix was data connection, not a new treatment — exactly inou's angle
+
+**Potential inou framing:**
+"This is why we built inou. Your labs, your inflammation markers, your genome — in one place. The connection no single practitioner has time to make."

memory/claude-usage.json

@@ -1,9 +1,9 @@
 {
-  "last_updated": "2026-03-15T04:00:01.760125Z",
+  "last_updated": "2026-03-15T10:00:01.976288Z",
   "source": "api",
-  "session_percent": 0,
-  "session_resets": null,
-  "weekly_percent": 13,
-  "weekly_resets": "2026-03-20T02:59:59.711524+00:00",
+  "session_percent": 3,
+  "session_resets": "2026-03-15T13:59:59.908041+00:00",
+  "weekly_percent": 14,
+  "weekly_resets": "2026-03-20T02:59:59.908059+00:00",
   "sonnet_percent": 16
 }

memory/updates/2026-03-15.json

@@ -0,0 +1,22 @@
+{
+  "date": "2026-03-15",
+  "timestamp": "2026-03-15T05:31:14-04:00",
+  "openclaw": {
+    "before": "OpenClaw 2026.3.13 (61d171a)",
+    "latest": "2026.3.13",
+    "after": "OpenClaw 2026.3.13 (61d171a)",
+    "updated": true
+  },
+  "claude_code": {
+    "before": "2.1.75",
+    "latest": "2.1.76",
+    "after": "2.1.75",
+    "updated": true
+  },
+  "os": {
+    "available": "0\n0",
+    "updated": false,
+    "packages": []
+  },
+  "gateway_restarted": true
+}

scripts/daily-updates.sh

@@ -53,27 +53,28 @@ echo "--- OpenClaw ---"
 OC_BEFORE=$(openclaw --version 2>/dev/null | head -1 || echo "unknown")
 OC_LATEST=$(npm show openclaw version 2>/dev/null || echo "unknown")
 echo "Current: $OC_BEFORE | Latest: $OC_LATEST"
-
 update_json "openclaw.before" "\"$OC_BEFORE\""
 update_json "openclaw.latest" "\"$OC_LATEST\""
 
-if [ "$OC_BEFORE" != "$OC_LATEST" ] && [ "$OC_LATEST" != "unknown" ]; then
-    echo "Updating OpenClaw..."
+# Use LLM to decide if these refer to the same version — no fragile string parsing
+OC_SAME=$(claude --print --permission-mode bypassPermissions "Do these two strings refer to the same software version? Answer only 'yes' or 'no'. String 1: '$OC_BEFORE' String 2: '$OC_LATEST'" 2>/dev/null | tr '[:upper:]' '[:lower:]' | tr -d '[:space:]')
+
+if [ "$OC_SAME" = "yes" ]; then
+    update_json "openclaw.updated" "false"
+    echo "Up to date: $OC_BEFORE"
+else
+    echo "Update available, running npm update -g openclaw..."
     if npm update -g openclaw 2>&1; then
         OC_AFTER=$(openclaw --version 2>/dev/null | head -1 || echo "unknown")
         update_json "openclaw.after" "\"$OC_AFTER\""
         update_json "openclaw.updated" "true"
         echo "Updated: $OC_BEFORE → $OC_AFTER"
-        # Reapply source patches
         /home/johan/clawd/scripts/openclaw-post-update-patches.sh 2>&1
     else
         update_json "openclaw.updated" "false"
         update_json "openclaw.error" "\"npm update failed\""
         echo "Update failed"
     fi
-else
-    update_json "openclaw.updated" "false"
-    echo "Up to date"
 fi
 
 # --- Claude Code ---
@@ -82,12 +83,16 @@ echo "--- Claude Code ---"
 CC_BEFORE=$(claude --version 2>/dev/null | sed 's/ (Claude Code)//' || echo "unknown")
 CC_LATEST=$(npm show @anthropic-ai/claude-code version 2>/dev/null || echo "unknown")
 echo "Current: $CC_BEFORE | Latest: $CC_LATEST"
-
 update_json "claude_code.before" "\"$CC_BEFORE\""
 update_json "claude_code.latest" "\"$CC_LATEST\""
 
-if [ "$CC_BEFORE" != "$CC_LATEST" ] && [ "$CC_LATEST" != "unknown" ]; then
-    echo "Updating Claude Code..."
+CC_SAME=$(claude --print --permission-mode bypassPermissions "Do these two strings refer to the same software version? Answer only 'yes' or 'no'. String 1: '$CC_BEFORE' String 2: '$CC_LATEST'" 2>/dev/null | tr '[:upper:]' '[:lower:]' | tr -d '[:space:]')
+
+if [ "$CC_SAME" = "yes" ]; then
+    update_json "claude_code.updated" "false"
+    echo "Up to date: $CC_BEFORE"
+else
+    echo "Update available, running npm update -g @anthropic-ai/claude-code..."
     if npm update -g @anthropic-ai/claude-code 2>&1; then
         CC_AFTER=$(claude --version 2>/dev/null | sed 's/ (Claude Code)//' || echo "unknown")
         update_json "claude_code.after" "\"$CC_AFTER\""
@@ -98,9 +103,6 @@ if [ "$CC_BEFORE" != "$CC_LATEST" ] && [ "$CC_LATEST" != "unknown" ]; then
         update_json "claude_code.error" "\"npm update failed\""
         echo "Update failed"
     fi
-else
-    update_json "claude_code.updated" "false"
-    echo "Up to date"
 fi
 
 # --- OS Packages ---