Claude Code Scheduled Tasks

Desktop-app-only feature that lets Claude Code run agents on a cron schedule with persistent storage and catchup. Unlike Loop, scheduled tasks survive across app restarts and fire up to 7 days of missed runs when the desktop app reopens. Each run is stateless with no shared memory, so self-improvement patterns typically write a per-task log file that the next run reads before executing.

Key points

• Only available in the Claude Code desktop app at launch
• Persistent across app restarts with 7-day catchup on missed runs
• Each run is stateless — no shared memory between runs by default
• Self-improvement pattern: single-file per-task log overwritten by each run
• Can be configured with name, description, prompt, model, mode, folder, and cron
• Uses the same agent harness as interactive Claude Code — agentic and self-healing
• Notifications via desktop hook (sound) or in-prompt ClickUp/Slack messages
• Nate's later deployment comparison distinguishes local scheduled tasks from cloud routines: local tasks keep local-machine context and can catch up when the app reopens, while cloud routines run without the machine but need cloud-safe repo/env setup ^[src-086].

Related entities

• Claude Code

Related concepts

• Claude Code Loop Skill
• Claude Code Memory 2.0 (AutoDream)
• Agent Deployment Modes

Source references

• ^[src-004] Nate Herk cluster — Nate Herk — Claude Code cluster (21 videos)

– Videos referenced: BlNJFa3Btm8, OUyfxhFtGCo

• ^[src-086] Nate Herk — "I Tested 3 Ways to Deploy Claude Agents (Here's When to Use Each)" (2026-05-15)