Schedulers & Automation
Scheduling is where the agent acts on time instead of on a prompt. In Hermes a cron job
is a first-class agent task, not a shell line: each run is a fresh AIAgent with no
history, given the skills it needs. That framing changes how you design jobs — they
must be idempotent, deliver somewhere, and stay quiet until they have something worth
saying.
Cron = time. Each job is a clean agent run with attached skills, so it behaves the same on run #1 and run #100. The discipline: idempotency, an explicit delivery target, alerts that signal instead of spam, and human approval before anything sensitive fires.
Planned notes
- Cron jobs as agent tasks, not shell jobs
- Each run = a fresh AIAgent with no history
- Attaching skills to a scheduled job
- One-shot vs recurring schedules
- Delivery targets: where a job's output lands
- Idempotency: safe to run twice
- Alerts vs noise: signaling without spamming
- Human approval for sensitive scheduled actions
- Candidate jobs: daily briefing, opportunity radar, freelance scan, inbox triage, GitHub review, atlas maintenance
- Observability for scheduled runs
Core sources
- Hermes — Feature: Cron — https://hermes-agent.nousresearch.com/docs/user-guide/features/cron
- Hermes — Features overview — https://hermes-agent.nousresearch.com/docs/user-guide/features/
- Hermes — Security (approval for sensitive actions) — https://hermes-agent.nousresearch.com/docs/user-guide/security
Connects to: Native Orchestration · Income Workflows · Agent Security & Ops · Evaluation & Observability