# The Holy Trinity: Data + AI + Automation

**Give your business a memory, a team and a rhythm.** Data grounds the work. AI produces and checks it. Automation repeats useful routines. You review the decisions.

| Element | Practical starting point | Result |
| --- | --- | --- |
| **Data** | Business brief, source documents, sanitised conversations | Claims you can trace to evidence. |
| **AI** | Clear roles, bounded tasks, handoff packets | Useful drafts with visible ownership and checks. |
| **Automation** | One tested routine on a daily or weekly schedule | Meaningful-change notifications and prepared decisions. |

## 1. Set up your workspace

Choose one command, then continue with steps 2-5.

**All-in-one:** workspace and assistant templates, plus optional MangoMagic installation, model registration and a graceful ChatGPT restart.

```bash
bash -c 'f=$(mktemp) || exit; trap "rm -f \"$f\"" EXIT; curl -fsSL https://raw.githubusercontent.com/mango-magic/mangomagic/main/setup-operations.sh -o "$f" && bash "$f" --with-mangomagic'
```

**Workspace and assistant templates only:** use your current model; add MangoMagic later in step 6 if wanted.

```bash
bash -c 'f=$(mktemp) || exit; trap "rm -f \"$f\"" EXIT; curl -fsSL https://raw.githubusercontent.com/mango-magic/mangomagic/main/setup-operations.sh -o "$f" && bash "$f"'
```

Both populate `~/Documents/AI Operations` with the starter files and preserve existing files. Neither completes account sign-in, project connection or your assistant interview, launches workers or creates schedules. Add `--no-restart` after `--with-mangomagic` to defer the ChatGPT restart.

## 2. Connect the project

**In the ChatGPT desktop app:** create a local project, open its menu → **Edit project → Add folder**, select the installed `AI Operations` folder and choose **Make primary**. Start a new chat there. The primary folder supplies the default working directory and automatic discovery of project instructions and configuration; merely attaching a secondary folder does not provide that discovery. [Official project setup](https://learn.chatgpt.com/docs/projects).

If using an ordinary cloud ChatGPT Project, upload the relevant documents and paste operating rules into project instructions. It does not automatically read local folders. Supply updated copies when files change; request labelled file contents if writing is unavailable.

## 3. Build a useful assistant

The onboarding pack adapts the Team AI Operations source. Read [START_HERE.md](../starter/START_HERE.md), then use [BUILD_MY_ASSISTANT.md](../starter/00_Command_Centre/BUILD_MY_ASSISTANT.md) with the existing [ASSISTANT_PROFILE.md](../starter/00_Command_Centre/ASSISTANT_PROFILE.md) and [ASSISTANT_ROLLOUT.md](../starter/00_Command_Centre/ASSISTANT_ROLLOUT.md).

**Paste into the connected project chat:**

```text
Build my assistant. Read START_HERE.md and
00_Command_Centre/BUILD_MY_ASSISTANT.md first. Verify that you can read the
required files; report missing access rather than pretending setup worked.

Use the existing project context, my previous answers and authorisation already
provided; do not ask for the same approval again. Take me through the
brief in the guide, asking only for material gaps. Preserve existing work and
mark unknowns. Capture my preferences from examples and feedback, without
personality labels. Use the existing assistant profile and rollout templates;
do not create a competing folder structure or task tracker.

Help me choose one useful first deliverable, agree its acceptance checks,
produce it and test it against the available evidence. Record the result,
limitations and next action in the existing files. Keep final review with me.
Account connections, external actions and schedules need their actual tools
and my authorisation; never report them active just because files exist.
```

### Know where the work lives

Paths below are relative to the installed folder. Use the existing files rather than creating another hierarchy.

| Location | Purpose |
| --- | --- |
| `START_HERE.md`, `AGENTS.md` | Onboarding and shared responsibilities, boundaries and working rules. |
| `CURRENT_MISSION.md`, `project_tasks.json` | Root mission and workspace task evidence, including reserved owner review. |
| `00_Command_Centre/BUSINESS_BRIEF.md` | Role, team priorities, source files, standards and permissions. |
| `00_Command_Centre/TASK_BRIEF.md`, `00_Command_Centre/HANDOFF.md` | Reusable task and delegation templates. |
| `01_Data/SOURCE_REGISTER.csv` | Source IDs, locations, dates, authority and permitted use. |
| `01_Data/source-documents/`, `01_Data/sanitised-conversations/` | Source library and privacy-safe conversation extracts. |
| `02_Agents/AGENT_ROSTER.md`, `02_Agents/ROLE_TEMPLATE.md` | Role mapping and template for additional responsibilities. |
| `02_Agents/<role>/AGENTS.md` | Each role's mission, inputs, outputs, boundaries and checks. |
| `.codex/agents/*.toml` | Six actual project-scoped custom agent definitions. |
| `03_Projects/project-template/` | Copy for each outcome; includes `TASK_BRIEF.md` and `project_tasks.json`. |
| `04_Automations/AUTOMATION_BRIEF.md` | Routine, schedule proposal, manual test and scheduling evidence. |
| `05_Deliverables/` | Links to reviewed final outputs. |

## 4. Work with your team

The six roles are chief-of-staff, sales, research, content, automation and quality. Their custom names are listed in the roster, such as `mango_sales` and `mango_quality`.

**Templates alone do not spawn agents.** This starter also includes TOML definitions in the documented project location, `.codex/agents/`. Supported local Codex clients can use them when delegating. Verify availability in a new project chat; opening the folder does not launch six workers. These definitions inherit parent settings rather than granting new access. [Official subagent configuration](https://learn.chatgpt.com/docs/agent-configuration/subagents).

Use the loop: **brief → produce → verify → review → repeat**.

- **Ground:** register sources and preserve originals. Use sanitised conversation extracts and traceable source IDs.
- **Assign:** copy the project template for one outcome. Fill its task brief with scope, owner, evidence, output location and acceptance checks. Explicitly request 3-5 specialists when useful and available; otherwise work sequentially.
- **Produce:** give each specialist a separate write scope and require the existing handoff format. Read relevant role instructions explicitly; nested role files are not all loaded from a root chat.
- **Verify:** check claims, calculations, privacy and acceptance criteria. Update the relevant `project_tasks.json` with actual status and evidence paths. Keep root tracking at workspace level and project detail in its project file.
- **Review:** present the completed draft, checks, gaps and decisions to the CEO. Keep the owner review task `reserved_for_owner` until review occurs. Link the approved project output from `05_Deliverables/` with reviewer and date.

**Evidence beats percentages.** A draft file proves a draft exists; completion requires acceptance evidence. Never invent progress or mark owner review complete on the owner's behalf.

## 5. Automate a tested routine

Try a **daily mission check** for new blockers or completed work, or a **weekly team review** for changes in priorities, delivery and blockers.

Fill the automation brief, test the routine manually, then create and verify the schedule in the app's scheduling UI. Check timezone, enabled status, input access and the first run's result. Local-file jobs need the computer on, the app running and the project available. **Files themselves do not schedule anything.**

Notify only on meaningful changes, completion, actionable failure or a required decision. Keep external actions within explicit authorisation and final review with the CEO.

## 6. Optional: add MangoMagic

If all-in-one setup succeeded, the model installation is already done. Otherwise, use the [optional MangoMagic step](../README.md#6-optional-add-mangomagic-71) when you want it. The workspace and assistant workflow can use your existing model.

Use [the copy/paste prompts](PROMPTS.md) for your next outcome.