Bot = role + stop rules in the description. Skill = reusable procedure. Routine = scheduled or event-triggered run of a job. Keep secrets out of all three. Routines fire on time (cron) or matching outside events your desk wires—not “whenever the Bot feels like it.” Approval boundaries live in the Bot description (and AGENTS.md for repo agents); skills and routines must not silently bypass stop-before-send/publish/pay. Don’t CreateAgent when a skill or a clearer description on an existing Bot covers the job. Soft path: /free, Cheat Sheet for the vocabulary checklist, then Agent Starter Kit ($47) when you want harness depth.
Who this is for
Solo founders and small ops desks already in Cursor + Grok Bot who need shared vocabulary before they add seats. Seat templates and hard stops: Grok Bot role templates that stop before send/publish/pay. Coordination pattern: Lead Bot + specialist Bot coordination. This post owns the three-layer split only.
Definitions (ops vocabulary)
Keep the words boring and portable:
- Bot (description) — who the seat is, what “done” means, and the hard stops (never send / publish / pay / delete / share outside the box without approval).
- Skill — a reusable procedure (draft a brief, assemble a checklist). Skills say *how*; they must not silently override a stop in the description.
- Routine — a scheduled or event-triggered run of a Bot + skill. Routines fire work; they do not grant permission to leave the box.
Secrets stay in env / secret store—not pasted into description, skill, or routine prompt text.
Hard boundaries (never auto-send / publish / pay)
Put these in the Bot description in plain language the model can’t “creatively reinterpret”:
- Send — email, Slack, SMS, chat: draft only; human Approves before send.
- Publish — blog, social, changelog, store listing: draft artifact; human hits publish.
- Pay — Stripe, invoices, refunds, marketplace buys: propose; human executes.
- Delete / share outside the box — trash, public share links, credential paste: stop and ask.
“Cold” work that *is* allowed: research, outlines, in-box drafts, checklists, local file edits the human reviews. If the action is irreversible or externally visible, the Bot’s job ends at a reviewable artifact.
When not to CreateAgent
| Situation | Do this | Not this |
|---|---|---|
| Same job, clearer how-to | Add or tighten a skill | Spawn a duplicate Bot personality |
| Same seat, fuzzy stops | Rewrite the Bot description stop list | CreateAgent “because it felt stuck” |
| Recurring draft work | Wire a routine that only starts drafts | Routine that auto-sends / publishes / pays |
| New craft lane (content vs ops) | Named seat + artifact type + stops | Mega-Bot that claims every job |
New seats need a named artifact and a hard stop before send/publish/pay—not a second personality with the same vague brief.
What’s “done” for each layer
| Layer | Done means |
|---|---|
| Bot | A reviewable artifact the human can approve, edit, or reject in one pass |
| Skill | A completed procedure (checklist filled, draft written) |
| Routine | The scheduled job finished and reported—or stayed quiet if its prompt says so |
Document the routine’s trigger in the routine itself, not only in chat. Time (cron) or matching outside events (e.g. chat/repo listeners) are valid; vibes are not.
Decision table
| If you need… | Use… | Never… |
|---|---|---|
| A seat with stop rules | Bot description | Putting stops only inside a skill |
| A reusable how-to | Skill | Overriding description stops |
| Work on a schedule or event | Routine | Treating the trigger as permission to send/publish/pay |
| A new craft lane | New Bot + artifact type | CreateAgent for the same vague job |
| Repo-wide verify / guardrails | `AGENTS.md` | Secrets in any chat surface |
Debug order: read Bot description stop list → confirm skill has no send/publish/pay side effects → confirm routine only triggers drafts → then add seats.
What Cheat Sheet + Agent Starter are for
The Cheat Sheet is the skimmable vocabulary and approval checklist for this desk—name it in your runbook; product URL ships when live. Soft path today: start on /free (Harness Desk Starter—seat, done-artifact skill, gate card). When you want harness depth (role pack + agent-ops files, not AI-SDR SaaS), use the Agent Starter Kit ($47).
Stay in agent-ops—never Domo-as-kit-ROI, never “replaces your team,” never a routine that auto-sends under your name.
Soft next step
Name Bot / skill / routine once, put never-send/publish/pay in every Bot description, and prove one routine ends as a draft—not a send. Soft ladder: /free for the starter desk, Cheat Sheet for the vocabulary checklist, then the Agent Starter Kit ($47) when you want harness depth. No rush—ship the three-layer card into your own runbook either way.