You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fully autonomous infrastructure that monitors, tests, screenshots, audits, and self-heals 12+ applications around the clock. Claude Code AI agents are deployed automatically to fix crashes and flag security/performance issues.
Always Running
Frequency
Job
Script
What It Does
Every 30s
Auto-Restart
server.js (built-in)
Detects any down app, restarts via launchctl instantly
Every 5 min
Health Check + Fix
health-check-fix.sh
Quick fixes first (kill stale, clear locks), then deploys one Claude Code agent per broken app to diagnose and repair
Nightly Pipeline (12 AM - 6 AM)
Time
Job
Script
What It Does
Auto-Fix?
12:00 AM
Git Pull
git-pull-all.sh
Pulls latest code from GitHub across all 12 repos before nightly jobs begin
N/A
1:00 AM
Test Suite
nightly-tests.sh
Runs 877+ unit/integration/e2e tests across 11 repos. Any failure triggers a Claude Code agent to fix, verify, commit, and push
Yes
2:00 AM
Screenshots
nightly-screenshots.sh
Retakes all app screenshots (desktop 1920x1080 @2x, mobile 390x844 @3x) with MacBook Air and iPhone frames
N/A
3:00 AM
GIF Recordings
nightly-gifs.sh
Records 30s of real page activity per page, compresses to 5s looping HD GIF. 25 pages for 3pi via LAN, 5 parallel batches
N/A
4:00 AM
Deep Audit
deep-audit.sh
Scans for Turbopack cache corruption, native module mismatches, port conflicts, stale lock files, disk space. Fixes automatically and restarts affected apps
Yes
5:00 AM
Security + Performance Scan
nightly-scan.sh
Deploys one Claude Code agent per app to audit for hardcoded secrets, XSS/SQL injection risks, large bundle imports, missing optimizations. Flags to stickies, does NOT auto-fix
No (flags only)
6:00 AM
Summary Report
nightly-summary.sh
Aggregates results from all nightly jobs and posts a combined report to stickies
N/A
Idle Hours (7 AM - 11:59 PM)
No scheduled jobs run during work hours. Only the always-running monitors (30s restart + 5 min health check) remain active.
Claude Code Agent Deployment
Trigger
Budget
What the Agent Does
Auto-Commit?
App down after quick fix fails (5 min check)
$1.00/agent
Reads crash logs, checks deps/ports/modules, fixes and restarts
Yes, if code changed
Test failure (1 AM)
$1.00/agent
Reads test output + source, fixes code or tests, verifies, pushes
Yes
Security/performance scan (5 AM)
$0.50/agent
Scans source for vulnerabilities and performance issues, reports flags
No (read-only)
Agent Guardrails
Maximum 3 fix attempts per agent
Cannot delete tests or add .skip
Cannot change application functionality
Must verify fix with npm test before committing
Restricted tools: Bash, Edit, Read, Write, Glob, Grep only
Security/performance agents are read-only (no writes)
Monitored Applications
App
Port
Path
Test Count
bheng
3000
/Users/bheng/Sites/bheng
172 (153 unit + 19 e2e)
tools
3001
/Users/bheng/Sites/tools
16
diagrams
3002
/Users/bheng/Sites/diagrams
10
claude
3003
/Users/bheng/Sites/claude
11
3pi
3333
/Users/bheng/Sites/3pi
10
3pi-poc
3334
/Users/bheng/Sites/3pi-poc
9
stickies
4444
/Users/bheng/Sites/stickies
266
vault
4445
/Users/bheng/Sites/vault
183
mindmaps
5173
/Users/bheng/Sites/mindmaps
82
safe
6100
/Users/bheng/Sites/safe
10
drop-web
3010
/Users/bheng/Sites/drop
9
drop (Electron)
4111
/Users/bheng/Sites/drop/electron
Process check
drop-menu (Electron)
N/A
/Users/bheng/Sites/drop/electron
Process check
ai-spinner (Electron)
N/A
/Users/bheng/Sites/streamdeck
Process check
local-apps
9876
/Users/bheng/Sites/local-apps
API health check
Total: 877+ tests across 11 repos
Log Files
Log
Source
Contents
/tmp/git-pull-all.log
12 AM
Git pull results per repo
/tmp/nightly-tests.log
1 AM
Test results + agent fix details
/tmp/nightly-fix-{app}.log
1 AM
Per-app agent output (only on failures)
/tmp/nightly-screenshots.log
2 AM
Screenshot bot output
/tmp/nightly-gifs.log
3 AM
GIF recording output
/tmp/deep-audit.log
4 AM
Infrastructure audit results
/tmp/nightly-scan.log
5 AM
Security/performance flags
/tmp/scan-{app}.log
5 AM
Per-app scan agent output
/tmp/nightly-summary.log
6 AM
Aggregated summary
/tmp/health-check-fix.log
Every 5 min
Health check + fix results
/tmp/health-fix-{app}.log
Every 5 min
Per-app fix agent output (only on failures)
JSON Summaries (for dashboard integration)
File
Updated
Fields
/tmp/nightly-tests-summary.json
1 AM
total, passed, failed, fixed
/tmp/deep-audit-summary.json
4 AM
fixed, issues
/tmp/nightly-scan-summary.json
5 AM
securityFlags, performanceFlags, total
/tmp/health-check-summary.json
Every 5 min
total, up, down, fixed
Managing Crons
# List all scheduled jobs
launchctl list | grep com.bheng
# Disable a specific job
launchctl unload ~/Library/LaunchAgents/com.bheng.{job-name}.plist
# Re-enable a specific job
launchctl load -w ~/Library/LaunchAgents/com.bheng.{job-name}.plist
# Run any job manually right now
bash ~/Sites/local-apps/scripts/{script-name}.sh
# Check latest results
cat /tmp/{log-name}.log | tail -50
cat /tmp/{summary-name}.json
Hub/Agent Roles
All nightly jobs and health check agents only run on the hub machine (Mac Mini). Agent machines (MacBook Pro) skip all scheduled jobs silently.
Configured in local-apps/machine-role.json:
{"role": "hub"} // Mac Mini - all bots enabled
{"role": "agent"} // MacBook Pro - status reporting only
