Day 7 · Full Automation and Diagnostics

Today’s Goal

Two things:

1. Create a daily scheduled task — have Hermes send you a morning digest at a fixed time
2. Get comfortable with the diagnostic commands — at minimum: hermes doctor, hermes status, hermes logs

Coming from OpenClaw? After finishing this chapter, head to Migrating from OpenClaw — it’s a standalone migration guide.

---

Prerequisites

• Days 1–6 complete
• For the scheduled task section, Day 3 (at least one IM gateway configured) is ideal — that way the morning digest can push to Telegram / Discord etc.

---

Part A · Scheduled tasks: let Hermes run on its own

Hermes’s built-in cronjob tool lets the agent schedule and run its own tasks. Conceptually similar to unix cron, but every trigger is an independent agent session (with its own tools, skills, and memory) — not a bare script.

Three ways to create a job

1. Natural language (easiest)

   Start Hermes and say:

   You: Every morning at 9 AM, scan Hacker News, pick 3 AI-related stories, and send them to my Telegram.

   Hermes recognizes this as a scheduling request, calls cronjob(action="create", ...), and tells you: job ID, schedule rule, and next trigger time.

2. Slash command inside a conversation

   You: /cron add "0 9 * * *" "Scan HN, pick 3 AI stories, send to Telegram"

   Syntax: /cron add <schedule> "<prompt>".

3. Terminal CLI

   hermes cron create "0 9 * * *" "Scan HN, pick 3 AI stories, send to Telegram"

Schedule format

All formats supported:

Format	Meaning
30m, 2h, 1d	Relative delay, one-shot (run once in 30 minutes)
every 30m, every 2h	Recurring every N minutes/hours/days
0 9 * * *	Standard cron: every day at 9:00 AM
0 */6 * * *	Every 6 hours
2026-05-01T09:00:00	ISO timestamp, one-shot at a specific moment

Managing scheduled tasks

hermes cron list                      # list all jobs
hermes cron status <job_id>           # details and last result for a specific job
hermes cron edit <job_id> --schedule "0 10 * * *"
hermes cron tick                      # immediately check for due jobs (for debugging)

Inside a conversation:

/cron list
/cron run <job_id>      # trigger immediately, don't wait for the next scheduled time
/cron pause <job_id>
/cron resume <job_id>
/cron remove <job_id>

Running with a skill

Load a skill (installed in Day 5) before the cron task runs:

You: Every morning at 9 AM, use the blogwatcher skill to summarize my RSS feeds and send to Telegram.

Hermes uses the skill as the agent’s leading prompt, then stacks this task on top — effectively putting a Day 5 skill into an unattended pipeline.

Cron jobs cannot create other cron jobs

A cron task cannot schedule another cron task — this prevents runaway recursion. For A-then-B sequencing, describe the order inside a skill, or use two independent cron jobs with time offsets.

Silent mode: don't notify if there's nothing to report

If a task run has nothing worth reporting, prefix the response with [SILENT]. The result won’t push to your IM, but it will appear in ~/.hermes/cron/output/ audit logs. Great for monitoring tasks: “only bother me if something is wrong.”

When does the scheduler check?

The gateway (the long-running process from Day 3) checks for due jobs every 60 seconds. If the gateway isn’t running, cron won’t trigger — before relying on scheduled tasks, confirm:

hermes gateway status      # confirm gateway is running

---

Part B · The diagnostic trio

hermes doctor — automated health check

When anything strange happens, this is the first command to run:

hermes doctor
hermes doctor --fix        # auto-fix anything it can

It checks: Python / Node / dependency versions, config file syntax, database integrity, provider auth validity, gateway status, skills directory permissions, and more.

hermes status — current system state

hermes status              # summary
hermes status --all        # full detail (safe to paste into an issue — auto-redacts secrets)
hermes status --deep       # with active probing (sends a test request to the provider — slightly slower)

Use --all when filing issues — it redacts API keys and other sensitive values to sk-*** automatically.

hermes logs — log queries

Logs live in ~/.hermes/logs/. hermes logs is the query interface:

Log	What’s inside
agent (default)	API calls, tool dispatch, session activity
errors	Warnings and errors only
gateway	Platform connections, webhook events

Common scenarios:

hermes logs -f                         # tail in real time
hermes logs errors --since 1h          # errors from the last hour
hermes logs --level WARNING --since 30m
hermes logs gateway -n 100             # last 100 lines of gateway logs
hermes logs --session 7e4f2b8c         # all logs for a specific session

Typical debugging flow:

Symptom: Telegram bot isn't responding again

1. hermes gateway status                  → is it still alive?
2. hermes status --deep                   → is Telegram auth still valid?
3. hermes logs gateway --since 30m -f     → last 30 minutes of gateway logs
4. See 401 → re-run hermes gateway setup to refresh the token
5. See rate limit → configure fallback_model (covered in Day 2)

---

Seven-day recap

At this point you should have:

Day	Capability
1	Running hermes locally, completed first conversation
2	Chose a provider, configured a key, sent a reasoning test
3	At least one IM (Telegram / Discord / Slack) can @-mention the bot
4	Used at least 2 built-in tools: web_search and terminal
5	Installed a first skill and triggered it in a conversation ✨
6	Know ~/.hermes/memories/, /compress, and hermes backup
7	Write cron jobs to let Hermes run autonomously; use doctor/status/logs to diagnose issues

---

What’s next

The 7 days cover the minimum working loop. Where to go deeper depends on your use case:

Direction	Where to learn
Migrating from OpenClaw	Migrating from OpenClaw on this site
Embed Hermes in your own software	Official Developer Guide
Write skills for your team	Day 5 advanced section + agentskills.io examples
Connect internal services via MCP	Use MCP with Hermes
Voice interaction / self-hosting	Voice Mode / Self-hosting
Read the official docs end-to-end	Official Learning Path

This site will keep adding “Field Recipes” — each article focused on a specific workflow: code review agent, content aggregation bot, on-call operations assistant… complementary to this tutorial series.

---

Troubleshooting

Cron job didn't run at the scheduled time

99% of the time the gateway isn’t running. Check:

hermes gateway status
# Not running → hermes gateway start
# Status says running but still not triggering → hermes cron tick to see if manual trigger works

Cron triggered but result wasn't sent to Telegram

Delivery channel wasn’t specified. When creating the job, be explicit:

You: Scan HN at 9 AM daily, send the result to Telegram (not local).

Or edit the job: hermes cron edit <job_id> --channel telegram.

hermes doctor --fix reports 'nothing to fix' but the problem remains

--fix only handles issues it recognizes. Check hermes logs errors --since 1h for the specific error, then bring it to GitHub issues or the official Discord — include hermes status --all output (auto-redacted).

Moving this Hermes setup to another machine

Use the Day 6 backup and restore flow:

# On the original machine
hermes backup -o hermes-all.zip

# On the new machine (Hermes already installed)
hermes import hermes-all.zip
hermes gateway install && hermes gateway start

Running lots of automation — how to save tokens?

Use fallback_model in cron jobs to route to a cheaper model (covered in Day 2), or specify model: kimi-for-coding inside a skill to override — let routine repetitive tasks use a cheaper model, and save flagship models for important conversations.