Safe desk coordination is a small roster with hard seats—not one mega-Bot that tries to be chief of staff, writer, and publisher. The Lead Bot owns prioritization, handoffs, and review of anything that would leave the box. Specialists wake for one job, one artifact type, and hard stops before send, publish, or pay. Lead prepares drafts; a human (or explicit send permission) presses send. Nobody auto-emails or posts under the founder’s name. Soft path: /free, Cheat Sheet for the roster 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 want Lead → specialist handoffs without “the bot sent something I can’t undo.” Role templates and stop language live in the companion Grok Bot role templates that stop before send/publish/pay. This post owns the coordination pattern: what Lead owns, when specialists wake, and the approval gates that keep work in-box until a human says go.
Roster pattern (Lead + specialists)
Keep seats few so jobs don’t collide:
| Seat | Owns | Produces | Never does alone |
|---|---|---|---|
| Lead Bot | Prioritization, scoped handoffs, review pack | Consolidated review queue for the human | Auto-send / publish / pay |
| Content specialist | Drafts and outlines | `draft.md` / outline artifacts | Publish or post |
| Ops specialist | Runbooks, gate cards, digests | Checklist / status files | Spend, refund, delete |
| Support specialist | Reply drafts | In-box reply artifacts | Send the email |
| Launch specialist (optional) | Launch brief + checklist sequence | Brief / checklist files | Publish or blast send |
Prefer this small roster over one overloaded description that claims CoS + writer + publisher. Coordination beats a mega-Bot.
When specialists wake
- Lead receives a scoped ask from the human (or a routine that only *starts* draft work).
- Lead breaks the ask into specialist jobs: one job, one artifact type, named stop list.
- Specialist produces an in-box artifact—not an outbound message.
- Lead assembles a single review pack; human Approves, edits, or rejects.
- Only after an explicit gate does send / publish / pay happen—and that press is human (or an explicitly granted send permission), never silent auto-fire.
Specialists do not widen the brief alone. Lead reassigns; specialists stay in the seat they were given.
Boundaries (no auto-send / publish / pay)
Put these in every Bot description in 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, buys: propose; human executes.
- Delete / share outside the box — stop and ask.
“Done” for Lead and specialists is a reviewable artifact the human can approve in one pass—not an open-ended chat that might fire a tool later. Routines may schedule work; they do not grant irreversible rights. Secrets stay in env / secret stores—never in Bot descriptions, skills, or chat.
For repo agents that share the same desk, mirror stop-before-send/publish/pay in root `AGENTS.md` so Cursor and Claude Code inherit the same contract (One AGENTS.md…).
Coordination failure card
| Symptom | Likely cause | Fix |
|---|---|---|
| Lead (or specialist) sent/posted without asking | Stops missing from Bot description | Put never-send/publish/pay in the description; skills cannot override |
| Specialist “helps” by emailing | Lead delegated an outcome, not an artifact | Job = file/PR/URL draft; send gate stays with human |
| Mega-Bot does everything poorly | One description claims CoS + writer + publisher | Split into Lead + named specialists |
| Specialist widens scope mid-run | No “Lead reassigns” rule | Specialist stays in seat; Lead re-scopes |
| Routine pays or refunds | Routine treated as permission | Routines schedule drafts only |
| Chat says “done” with nothing to review | Done = conversation | Redefine done as a reviewable artifact |
Debug order: read Lead stop list → read specialist job + artifact type → confirm no send/publish/pay side effects in skills → confirm routine only triggers drafts → then add seats.
What Cheat Sheet + Agent Starter are for
The Cheat Sheet is the skimmable roster and approval checklist for Lead + specialist coordination—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 free Operator Kit fluff that still auto-sends under your name. Companion pillar for seat templates: Grok Bot role templates….
Soft next step
Prove one Lead → specialist handoff that ends as a reviewable artifact—not a send. Soft ladder: /free for the starter desk, Cheat Sheet for the roster/checklist mindset, then the Agent Starter Kit ($47) when you want harness depth. No rush—ship the coordination failure card into your own runbook either way.