Auto-Repair GitHub Issues with AI

Deploy issue-driven closed-loop automation for GitHub repos with cron-based agent fixes and delayed verification

Service/ops issues on GitHub that can be auto-fixed (config updates, restart cycles, cleanup) with delayed human verification.

• · GitHub repository OWNER/REPO
• · Issues labeled with type:service (or custom label)
• · Issue body describing the problem + optional ```verify``` code block for validation
• · Claude Code execution environment with git + gh CLI access

• · Auto-fixed issues with FIXED: marker
• · Verified fixes marked pending-verify → auto-closed on success
• · Failed fixes escalated to needs-decision + Telegram notification after N retries

• · GitHub CLI (gh)
• · Claude Code (cc) CLI
• · Python 3
• · ACL / setfacl (optional, for permission management)
• · cron (system scheduler)
• · Telegram API (optional, for notifications)

• · GitHub repository with CI/CD permissions for automated commits
• · Claude Code CLI installed and authenticated
• · Cron access on deployment host (usually /etc/cron.d/ or user crontab)
• · Issues properly formatted with type:service label and clear problem description

• · Verify block syntax error causes silent failure (code block must be valid shell)
• · Cron runs as wrong user (permissions insufficient for git push)
• · CC execution timeout on slow fixes (default likely 5-10 min per attempt)
• · Retry exhaustion without human escalation if Telegram webhook fails
• · Race condition if same issue processed in parallel cron runs

• · Full pipeline documented: issue routing → auto-fix → verify → close
• · Configurable cron interval, max retries, and logging destinations
• · Gradual escalation (pending-verify → needs-decision) prevents silent failures
• · Installation script handles user, permissions, and cron setup