> ## Documentation Index
> Fetch the complete documentation index at: https://engineering.datarelay.run/llms.txt
> Use this file to discover all available pages before exploring further.

# Quick Start

> The shortest path from a normal repository to the ChatGPT + Cursor + GitHub Engineering System workflow.

# Quick start

You do not need to adopt every optional tool on day one.

Start with the smallest loop that creates durable state:

```mermaid theme={null}
flowchart LR
    A["ChatGPT<br/>plan"] --> B["GitHub AI Work Packet"]
    B --> C["Cursor<br/>implement"]
    C --> D["Affected Tests"]
    D --> E["GitHub<br/>commit / PR / CI"]
    E --> A
```

## 0. Make GitHub reachable

Session continuity depends on both sides being able to read and update repository-scoped GitHub Issues.

* ChatGPT needs GitHub repository access for Work Packet creation/update.
* Cursor `/resume` needs an available GitHub integration or authenticated `gh`.
* If neither is available, stop and configure access instead of rebuilding state from pasted historical handoffs.

## 1. Add the repository entrypoint

A base Engineering System adoption needs:

```text theme={null}
AGENTS.md
.engineering/project.yaml
.engineering/tests.yaml
.engineering/release.yaml
```

Use the canonical templates from:

[github.com/datarelay-labs/engineering-system](https://github.com/datarelay-labs/engineering-system)

Pin the Engineering System version in the repository instead of silently following an unknown future version.

## 2. Add session continuity only if you use it

For ChatGPT ↔ Cursor resume workflow, add:

```text theme={null}
.cursor/commands/resume.md
.github/ISSUE_TEMPLATE/ai-work-packet.md
```

The optional repository Cursor rule can also point agents at the canonical Engineering System.

## 3. Keep ChatGPT instructions short

ChatGPT does not need a copy of the entire standard in Custom Instructions.

The global instruction only needs to route work correctly:

* identify the target repository first
* use the canonical Engineering System
* load minimal task-relevant context
* use the repository Work Packet for handoff/resume
* verify current Git/GitHub state before acting
* never claim unexecuted or different-HEAD evidence as PASS

## 4. Run the first workflow

```text theme={null}
ChatGPT:
"Cursor에 넘겨줘"

Cursor:
/resume

ChatGPT:
"Cursor 끝났어. 계속"
```

If the same workstream continues without a giant pasted prompt, the continuity path is working.

## 5. Add optional layers when they solve a real problem

<CardGroup cols={2}>
  <Card title="Desktop Commander" icon="terminal" href="/remote-audit">
    Add direct development-server audit when GitHub alone cannot prove local runtime or worktree state.
  </Card>

  <Card title="Athena" icon="book" href="/knowledge">
    Add searchable decision/history knowledge when rationale must be recovered across projects.
  </Card>

  <Card title="tmux + Phone SSH" icon="terminal" href="/mobile-remote-workflow">
    Add a persistent remote execution path when work must continue away from the desk.
  </Card>

  <Card title="Telegram" icon="bell">
    Add completion notifications when long-running implementation or tests should run unattended.
  </Card>
</CardGroup>

## Before implementation

The owner and AI should know:

```text theme={null}
change type
affected domain
public contract
persisted state
security boundary
operational impact
required tests
```

Then make the smallest correct change and run the cheapest deterministic check that can disprove it.

<Note>
  The owner retains product scope, final decisions, release approval, and human UX judgment. AI assists the engineering process; it does not silently broaden product scope.
</Note>