← All products
Pack

Agentic Loops Pack

The missing layer between a scheduled prompt and an agent that learns.

Any cron-based agent system — Hermes Agent, Claude Code, GitHub Actions, Cloudflare Workers, or any setup where an LLM runs a prompt on a schedule.
// what it is

A scheduled agent without state is a cron job with a language model attached. It scans the same sources, scores the same items, and hands you the same briefing it handed you yesterday. The fix is not a better prompt — it is a state file the agent reads on the way in and writes on the way out.

This pack is five such patterns, each with the YAML template and the JSON state schema that goes with it: delta detection for market and price scanners, carry-forward for briefings and to-do agents, topic lifecycle for research scanners, signal evolution for trend tracking, and baseline diff for security and health checks.

CRON-INTEGRATION.md covers wiring state reads and writes into an agent prompt you already have. COLD-START.md covers the first run, when there is no history to read yet — the case most stateful designs forget about.

// why it works
  • Delta detection — compare today against yesterday, report only what changed
  • Carry-forward — track items across runs and escalate anything stale for 3 days
  • Topic lifecycle — new → tracking → ongoing → mature, auto-pruning at 7+ days
  • Signal evolution and baseline diff — score histories, and new/resolved/changed/stable markers

Agentic Loops Pack

Turn your AI agents from re-prompts into self-improving loops. 5 stateful loop patterns + 5 JSON schemas + cron integration playbook.

Checkout and delivery handled by Gumroad. Instant download.