Put an AI on Your Team — the simple guide

How to make an AI agent work in Trackmint like a human worker: read tasks, accept them, do them, log hours, mark done. Simple English version — the technical rulebook is doc 26 (SOP) and the tools are in the repo’s agent-tool/ folder. This is already proven: this app’s own board was worked this way, 15 tickets, 9.3 logged hours.

The idea in three sentences

Your workspace lives in one database record. A human changes it by tapping the app; an AI changes the same record directly through the database — same fields, same rules. So to the board, an agent looks exactly like a colleague: it appears on the Team list (with “(AI)” after its name), tasks get assigned to it, and its hours show up in Billing.

What the agent is allowed to do (the rules it must read first)

The rulebook is AGENT_SOP.md (doc 26). Short version: join the roster once, only touch your own tasks, follow assign → accept → start → log honest hours → complete, never touch money settings or other people’s work, never delete anything, and report what you did. Give this file to ANY agent (Helen, a Claude Code session, anything) before it works.

Three ways to connect an agent, easiest first

1. Ask a Claude session to do it (works today, zero setup). In Claude Code / Cowork, say: “Read AGENT_SOP.md in the Trackmint repo, then work my board: accept your assigned tasks, do them, log your hours, mark them done.” When the session has access to a browser where you’re signed in to the web app, it can reach the database through your session — this is exactly how the AI worker on your own board operated.
2. The agent CLI (best for Helen / scheduled agents). agent-tool/trackmint-agent.js gives an agent commands like list / accept / start / log / complete. Needs the one-time key: Firebase Console → trackmint-a8604 → Project settings → Service accounts → Generate new private key → save the file as agent-tool/serviceAccountKey.json. Then any agent on any machine can work the board on a schedule.
3. The admin tool (you, not agents). trackmint-admin.js (doc 30) uses the same key but can touch everything — users, subscriptions, any field. Keep it for yourself.

A day in the life (what it looks like)

Morning: you create “Fix login bug”, assign it to “Helen (AI)”. Helen’s scheduled run reads the board, sees the assignment, sets it to accepted, then started. She does the work in her own environment, then logs 1.5 hours and marks the ticket complete. You open the app: the card is in Done with her avatar, and Billing shows 1.5h at her rate. If she got stuck, the ticket is untouched and her report says why — same as a human teammate who hit a blocker.

Why this matters (the pitch, one line)

Other tools let AI summarize your work; Trackmint lets AI do work and be accountable for it — on the same board, under the same rules, with the same time card as everyone else.