Automation Cron

Automation via Cron

I can schedule recurring tasks via the cronjob tool. They run as fresh agent sessions on the gateway scheduler, with results delivered back to the originating chat or to configured platforms.

Schedule formats

Hermes accepts both natural language and crontab syntax:

"every 1h"
"every 30m"
"every 2h"
"every 1d at 09:00"
"daily at 04:00"
"every monday at 09:00"
"0 9 * * *"           # crontab -- daily at 09:00 UTC
"30 */4 * * *"        # crontab -- every 4 hours at :30

Creating a job

Three surfaces, same effect:

# CLI
agent-machines cron create "every 1h" "Summarize new feed items" --skill blogwatcher --name "Hourly feeds"

# In-chat slash command
/cron add "0 9 * * *" "Audit my open PRs and post a summary"

# In a tool call
cronjob(
    action="create",
    schedule="every 1d at 09:00",
    prompt="Summarize my last 24 hours of commits across all projects",
    name="Daily commit digest",
    skills=["git-workflow"],
)

Skill-backed jobs

A cron job can preload one or more skills before running its prompt. Skills are loaded in order; the prompt becomes the task instruction layered on top.

Use this when the job needs a recurring procedure (e.g. blogwatcher checks feeds, llm-wiki updates a knowledge base).

Workdir scoping

By default cron jobs run detached from any repo. Pass workdir to load AGENTS.md / CLAUDE.md and run terminal/file tools rooted in that directory:

cronjob(
    action="create",
    schedule="every 1d at 09:00",
    workdir="/home/machine/projects/dedalus-monorepo",
    prompt="Audit open PRs, summarize CI health, and post to #eng",
)

The path must be absolute and exist; relative paths are rejected.

Operations

cronjob action=list          # show all jobs
cronjob action=pause id=...  # pause a job
cronjob action=resume id=... # resume a job
cronjob action=run id=...    # trigger immediately
cronjob action=remove id=... # delete a job
cronjob action=update id=... # edit schedule, prompt, skills, or workdir

Recursion guard

Cron-run sessions cannot themselves create more cron jobs. Hermes disables cronjob inside cron executions to prevent runaway scheduling loops. If you want a job to spawn other jobs, do it via direct tool calls in chat instead.

Patterns I run by default

These ship with the deployment and run automatically:

• Hourly health check -- verify the API server is responsive, restart the gateway if not.
• Daily wiki digest -- summarize what changed in ~/Documents/GitHub/my-wiki and post to chat.
• Weekly skill audit -- review my own skills folder, flag stale or duplicated entries.
• Nightly memory consolidation -- fold redundant entries in MEMORY.md, prune to fit the character limit.

When I propose a new cron

If the user describes a task that would benefit from recurrence ("every morning, check X"), I propose a cron with the exact cronjob action=create call before scheduling, so they can approve the schedule and prompt. Never add a cron silently.