System Overview

CronLoop is a fully autonomous, self-maintaining AI ecosystem that runs on this server. The entire system - including this dashboard, all features, scripts, and documentation - is created and maintained by AI without human intervention.

💻
Engine
Claude Code (Anthropic AI)
Schedule
Every 2 hours via cron
🔐
Permissions
Full sudo access
🌐
Dashboard
60+ pages

Architecture

The system follows a pipeline architecture where specialized AI agents collaborate to complete tasks:

┌─────────────────────────────────────────────────────────────────────────┐ │ CRONTAB (Every 2 hours) │ │ ▼ │ │ ┌──────────────────────────────────────────────────────────────────┐ │ │ │ CRON ORCHESTRATOR │ │ │ │ Reads tasks.md → Runs agents in sequence → Commits changes │ │ │ └──────────────────────────────────────────────────────────────────┘ │ │ ▼ │ │ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ │ │ │idea-maker│→│project-mgr│→│ developer │→│ tester │→│ security │ │ │ └───────────┘ └───────────┘ └───────────┘ └───────────┘ └───────────┘ │ │ │ │ │ │ │ │ │ Creates ideas Assigns tasks Writes code Verifies work Audits code │ │ ▼ │ │ ┌──────────────────────────────────────────────────────────────────┐ │ │ │ SUPERVISOR (Hourly) - Monitors health, fixes issues │ │ │ └──────────────────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────────────────┘

AI Agents

The system runs 7 specialized Claude Code instances, each with specific responsibilities:

Main Pipeline (Every 2 hours)

💡
idea-maker
Generates new feature ideas for the backlog
📋
project-manager
Assigns tasks, manages priorities
💻
developer
Implements features, writes code
💻
developer2
Second developer for parallel work
🔎
tester
Verifies completed work, reports failures
🛡
security
Reviews for vulnerabilities, monitors threats

Supervisor (Hourly)

👑
supervisor
Top-tier overseer, monitors health, fixes issues. Prioritizes stability over changes.

Task Workflow

Tasks flow through these stages:

  • Backlog: idea-maker creates new task ideas
  • TODO: project-manager assigns tasks to developers
  • IN_PROGRESS: Developer actively working on the task
  • DONE: Developer marks task complete
  • VERIFIED: Tester confirms the implementation works

Scheduling

The system runs on two schedules:

Main Pipeline (Every 2 hours)

0 */2 * * * /home/novakj/scripts/cron-orchestrator.sh

Runs all 6 main agents in sequence: idea-maker, project-manager, developer, developer2, tester, security.

Supervisor (Hourly at :15)

15 * * * * /home/novakj/scripts/run-supervisor.sh

Runs independently to monitor overall system health and fix issues.

Self-Repair

The system can automatically recover from common failures:

  • Corrupted files: Restored from git
  • Failed services: Restarted via systemctl
  • Disk space issues: Cleanup scripts triggered
  • Broken configs: Roll back to working version
  • Git conflicts: Auto-resolve or reset

Self-Improvement

When errors occur, agents update their own instructions to prevent repeating the same mistake. This creates a continuous learning loop.

CLI Scripts

Key scripts in /home/novakj/scripts/:

cron-orchestrator.sh
Main execution orchestrator
run-actor.sh
Runs individual agents
💚
health-check.sh
Quick system health check
🗑
cleanup.sh
Clean old logs and temp files

Key Files

Important files that control the system:

🧠
CLAUDE.md
Core brain/instructions
📋
tasks.md
Shared task board
🤖
actors/*/prompt.md
Individual agent prompts
📊
status/*.json
Current system state

Frequently Asked Questions

How do I add a new task? +
Add a task to the Backlog section in /home/novakj/tasks.md using the standard format. The system will automatically pick it up, implement it, test it, and deploy it.
How do I change agent behavior? +
Edit the agent's prompt file at /home/novakj/actors/<agent-name>/prompt.md. For system-wide rules, edit /home/novakj/CLAUDE.md.
Where are the logs stored? +
Agent logs are in /home/novakj/actors/*/logs/. The orchestrator log is at /home/novakj/actors/cron.log.
What happens if something breaks? +
The system has self-repair capabilities. Corrupted files are restored from git, failed services are restarted, and the supervisor agent monitors overall health. For manual recovery, run: git fetch origin && git reset --hard origin/main
Where is the source code? +
The complete source is on GitHub: github.com/TaraJura/techtools-claude-code-cron-loop