Workflow map

User guide › Workflow map

Start from what you want to do, follow it to the right scenario, and see the shape every scenario shares.

1. What do I want to do?

Find your task on the left, follow the arrows to a command. Open any command’s scenario page from the scenarios index.

flowchart TB
    Work(["New work <br>or a question"]) --> Q{"What do I <br>want to do?"}

    Q -->|"set up a repo for AI"| Init["/init-workspace-flow <br>(or ask in plain language)"]
    Q -->|"not sure"| Help["/help-flow"]

    Q ==>|"build or change something"| Build{{"Build / change"}}
    Build -->|"I know what to change"| Coding["/coding-flow"]
    Build -->|"pin down behavior first"| Reqs["/requirements-authoring-flow"]
    Build -->|"small or unusual chore"| Adhoc["/adhoc-flow"]

    Q ==>|"understand something"| Understand{{"Understand"}}
    Understand -->|"this codebase"| Analysis["/code-analysis-flow"]
    Understand -->|"options or technology"| Research["/research-flow"]

    Q ==>|"test something"| Test{{"Test / QA"}}
    Test -->|"design cases from a ticket"| TestGen["/testgen-flow"]
    Test -->|"automate a UI test"| UI["/ui-aqa-flow"]
    Test -->|"automate an API test"| API["/api-aqa-flow"]

    Q ==>|"transform or migrate"| Transform{{"Transform"}}
    Transform -->|"migrate or upgrade"| Modern["/modernization-flow"]
    Transform -->|"teach the agent a library"| ExtLib["/external-lib-flow"]

    Q ==>|"govern quality"| Govern{{"Govern"}}
    Govern -->|"security review"| Sec["/security-flow"]
    Govern -->|"author agent prompts"| Prompt["/coding-agents-prompting-flow"]

    classDef cmd fill:#dae8ff,stroke:#3674b5,color:#102a43,stroke-width:1.5px;
    classDef cat fill:#f4ddff,stroke:#884ea0,color:#351044,stroke-width:1.5px;
    classDef entry fill:#e5f7e8,stroke:#35834a,color:#153b20;
    classDef decide fill:#fff2cc,stroke:#b58b00,color:#493800,stroke-width:1.5px;
    class Init,Help,Coding,Reqs,Adhoc,Analysis,Research,TestGen,UI,API,Modern,ExtLib,Sec,Prompt cmd;
    class Build,Understand,Test,Transform,Govern cat;
    class Work entry;
    class Q decide;

Legend:

• 🟢 Green — where you start
• 🟡 Amber — a decision / your approval
• 🟣 Purple — category
• 🔵 Blue — a command

Diagram key on scenario pages: blue = a step the agent performs · amber diamond = a decision or your approval · red = a hard stop · green = start / finished. Arrows that curve back are feedback loops.

Tip: if the diagram doesn’t route you cleanly, just ask your agent /help-flow What should I use to … and it will point you to the right scenario.

2. How the process works

Whichever scenario you pick, the work follows the same five phases — Prepare → Research → Plan → Act → Validate — with approval gates where your judgment matters. Prepare happens once when you set up the repo; the rest repeats for every task.

flowchart LR
    Prep["Prepare <br><small>once, at setup</small>"] --> Research["Research <br>gather context"]
    Research --> Plan["Plan <br>specs &amp; approach"]
    Plan --> G1{"You approve <br>the plan"}
    G1 -->|approve| Act["Act <br>implement"]
    Act --> Validate["Validate <br>review &amp; actually run it"]
    Validate --> G2{"You approve <br>the result"}
    G2 -->|approve| Done(["Shipped"])
    G2 -->|issues found| Research

    classDef step fill:#dae8ff,stroke:#3674b5,color:#102a43;
    classDef gate fill:#fff2cc,stroke:#b58b00,color:#493800;
    classDef done fill:#e5f7e8,stroke:#35834a,color:#153b20;
    class Prep,Research,Plan,Act,Validate step;
    class G1,G2 gate;
    class Done done;

Bigger tasks add independent reviewer and validator subagents and more gates; small tasks collapse the gates and skip the heavy checks. The agent won’t move past an amber gate without a clear go-ahead from you.

Part of the Rosetta user guide. New here? Start with What is Rosetta? · Install · Set up your repo.