Lesson 6 · OpenClaw, Safely Pro ~11 min read 4 starter workflows

Your first real workflows: briefing, triage, watchers.

A safe agent that does nothing useful is a smoke detector with no kitchen. These four starter workflows are chosen on two criteria: they deliver real daily value, and every one stays inside the safety rails you built in Lessons 3–5. Notice what's not here — nothing that needs your passwords, your main inbox, or marketplace skills.

01 The morning briefing — your agent's signature move

1

One message, waiting for you at 7am

The classic always-on win: a scheduled job that assembles your morning in one chat message. Weather, your calendar (the agent's copy — see the pattern below), top headlines on topics you care about, and anything your watchers (workflow 3) flagged overnight.

Tell your agent (in your own words)

Setting it up conversationallyEvery weekday at 7:00am, send me a briefing: 1. Weather for Fargo today, one line. 2. My calendar events for today (from the shared calendar). 3. Top 3 news items on: AI tools, small business tech. One sentence each, with links. 4. Anything your watchers flagged since yesterday. Keep the whole thing under 200 words. If something failed overnight, say so first.

OpenClaw turns this into a cron job (Lesson 9 goes deep on those). The "if something failed, say so first" line teaches your agent the most underrated habit: reporting its own failures instead of silently skipping.

Value: your day's dashboard, in the app you check first anyway.

02 Email triage — on the agent's OWN inbox

2

The dedicated-inbox pattern (this is the important part)

Remember Lesson 3: the agent gets its own accounts. So don't wire it into your personal Gmail — instead, make the agent's inbox useful by forwarding into it. Set rules in YOUR email that forward selected, low-risk streams: newsletters, receipts, shipping notices, school announcements, that one chatty mailing list. The agent reads only what you've chosen to send it.

Then ask for triage like this

Triage instructionEach evening at 6pm, summarize what arrived in your inbox today: - Anything that needs action from me (deadlines, RSVPs, payments due) — list first, bold. - Receipts and orders: one line each. - Newsletters: just the 3 most interesting headlines total. Never click links in emails. Never reply to anyone. Read and report only.

"Never click links, never reply" matters: emails are exactly where prompt-injection payloads arrive (Lesson 2). A read-and-report agent on a forwarded-mail diet gets you 80% of the value at a tiny fraction of the risk of full inbox access.

Value: newsletter pile → 60-second digest, with deadlines surfaced.

03 Watchers — the patience you don't have

3

Check a thing, every day, forever

Anything you'd check repeatedly is a watcher: a price you're waiting to drop, a permit status page, concert tickets, a competitor's pricing page, the release notes of software you depend on. The agent browses on schedule, compares against last time, and messages you only when something changed.

Watcher patternOnce a day, check [URL]. Compare to what you saw yesterday. If [the price drops below $X / the status changes / new content appears], message me immediately with what changed. If nothing changed, stay silent — no daily "no news" messages. If the page won't load two days in a row, tell me that too.

"Stay silent unless something changed" is the difference between an assistant and a notification problem. Three or four watchers running quietly is where an always-on agent starts feeling like a superpower.

Value: you stop checking; it never forgets to.

04 Research errands — fire and forget

4

Send a question from the grocery line, read the answer at your desk

The ambient win: any moment a question occurs to you, hand it off in one message. The agent researches with the time you don't have and files the answer back — in chat, or into a notes file in its workspace folder you sync later.

Errand brief (the Foundations patterns apply!)Research errand, no rush — reply within the hour: Best options for [X] under [$Y]. Compare the top 3 on [criteria]. Cite sources with links. If reviews disagree, say so instead of averaging. Save the full version to notes/research/ and send me just the 5-line summary.

Notice this is just a Foundations-style brief with a delivery instruction. Everything you learned about constraint stacks and hallucination flags applies to your agent — it's the same discipline, with the answers arriving in your pocket.

All four workflows are read-and-report. None of them let the agent send email to other people, spend money, post publicly, or touch files outside its workspace. That's deliberate. An agent earns "write access" to your life the way a new employee earns it: weeks of clean, observable work first — and even then, one scope at a time (Lesson 8 covers exactly how).
The cost check-in

Four daily workflows on a mid-tier model typically lands in the $5–15/month range depending on how chatty your briefings are. Check the usage dashboard after week one. If costs surprise you, the briefing word limits and "stay silent" rules above are the dials that matter.

Build two this week

The morning briefing plus one watcher is the perfect starter pair — one scheduled push, one quiet monitor. Run them a full week. When both have been boringly reliable for seven days, you've earned the next power-up: skills. Which is also where the danger comes back — so don't skip the vetting lesson.

What you can do now

  • Set up a scheduled morning briefing with failure-reporting built in
  • Run email triage on a dedicated, forward-fed inbox — read-and-report only, never clicking links
  • Deploy watchers that alert on change and stay silent otherwise
  • Fire off research errands as proper briefs from anywhere
  • Explain why all starter workflows are read-only — and what an agent must do to earn write access
  • Estimate and tune your monthly model spend
Pro
Up next in OpenClaw, Safely

Lesson 7 · Skills: the app store with a malware problem

ClawHub has thousands of skills — and an audit found roughly a third with prompt-injection content. The vetting checklist, and why the safest skill is the one you write yourself. See pricing →