Field notes. John Arndt, Soxoa. Published .
The usual way AI work goes wrong for a business owner isn't a dramatic failure. It's quieter than that. The agent finishes, reports success, and three things are still waiting on the owner. Nobody wrote them down. The payment webhook isn't on. The new sender domain isn't verified. A decision about the free tier is sitting in paragraph nine of a long reply. A week later something is broken, and it turns out it was never set up.
owner-actions exists so that every piece of agent work ends with one clear handoff.
Where the line is
Some things the agent should do itself: run commands, edit files, test, draft, verify. Others belong to the owner, every time: signing in, granting access, approving multi-factor prompts; anything that spends money; anything sent to customers, prospects or the public; settings in third-party dashboards: billing, DNS, security; legal and compliance sign-off; decisions about what to build, stop or charge; and rotating any password or key that appeared in a chat.
The rule runs both ways. If the agent can do a step, it does it instead of listing it. The owner's list should only hold things that need the owner.
What a good item looks like
Each item is something you can act on in a minute. It gives the exact place (a URL or menu path), the exact value to paste, one line on why, and what happens if it doesn't get done:
“2. Turn on the payment webhook. Dashboard → Developers → Webhooks → Add endpoint → paste the address → select "checkout completed". Why: paid orders aren't being recorded. If skipped: every sale needs manual entry.”
Items are numbered across the whole list, so you can reply "did 2 and 4" and be understood. They're grouped by urgency: do now (money or customers affected), toggles to flip when ready, and decisions.
Done means proven
The top section lists what the agent finished, and every line carries proof: a test result, a live link, a commit. Anything without proof goes on a "done, not verified" line instead. That's what lets you trust the rest of the list.
Two rules learned the hard way
Never write a secret into the list. Name it, say where it goes, and let the owner paste it. A handoff document full of passwords is now a document that needs protecting.
Name intentional gates. A human review queue that's holding work is doing its job. It gets listed as a gate so nobody "fixes" it.
Unfinished items carry forward to the next dated list. Nothing gets dropped without a word.
Get the skill
owner-actions is part of the free Soxoa Ops plugin for Claude. In Claude Code, add the marketplace with /plugin marketplace add johnarndt/soxoa-claude-plugins, then run /plugin install soxoa-ops@soxoa-plugins. You can read the skill on GitHub before installing it, or copy its folder into ~/.claude/skills/.
Deciding where the human sits in each workflow is most of what a good AI rollout is. Here's how we approach it.