Workflows

A single agent loop decides its own path: it reads, edits, runs and answers in whatever order it chooses. A workflow decides the path for the agent. It says which steps run, in what order, what each step may use, and what must be checked before it can finish.

Rusty ships 15 workflows, available in every project, from a two-step investigation to a full audit, fix and re-audit. You pick one from the Mode menu in Agent Mode, or let Rusty choose for you.

	Single agent	Workflow
Path	The agent decides.	The workflow decides.
Tools	Whatever the chat's skill allows.	Each step has its own: a planning step cannot edit files at all.
Context	One long conversation.	Each step starts fresh and receives the previous step's written handoff.
Checks	Up to the agent.	Completion gates: Build cannot report until it has checked its own edits.
Best for	Quick questions and small edits.	Work you want broken into steps you can follow, and review.

Run a workflow

1. Open Agent Mode.
2. In the workflow bar above the message box, open the Mode menu and choose a workflow. Stages are labelled Stage:, pipelines Workflow:. Hover the summary line to read what the selected one is for.
3. Describe what you want and press Enter.

While it runs, the bar lists the steps in order and tells you where it is, for example Running step 2 of 4: Build. A step waiting on a command prompt says so, and a step that is retrying shows its attempt. Each agent step runs in a fresh, isolated session that inherits your workspace, your model and your permission policies, and the chat marks where each step's output begins.

The workflow's final message is its last step's result, so read it. A finished run means the sequence finished, not that a review approved the change.

If a step fails, the bar says so and your next message resumes at the failed step. Steps that already finished are kept. Choose Start over instead to discard them.

Stages and pipelines

The built-in workflows come in two kinds.

Stages are short runs of two steps, almost all of them read-only. Each stops at a written result you can read and amend: findings, a design, a plan, a verdict.

Pipelines run a whole job from start to finish: plan, build, verify, and so on.

Use a pipeline when the path is clear and you want it done. Use stages when you want to look at each result before the next thing happens, when you are not sure yet what to build, or when the change is risky.

See Built-in workflows for every one, step by step.

Steer a long job between stages

A stage ends with a result in the chat. The next workflow you run in the same chat starts from that result: Rusty hands it to the first step as context, and that step builds on it instead of starting over.

That gives you a point to intervene between stages. For example:

1. Run Stage: Research & analyze on "how does sync work". Read the findings.
2. Reply with your correction, such as "keep the queue, skip the migration, and design the retry fix", and run Stage: Architect & plan. It sees the findings and your correction. Read the plan and edit it in your next message if you want.
3. Run Stage: Build & verify to carry the plan out, then Stage: Verify & review changes for a second look.

Things to know:

• The context is the last result in the chat: the most recent assistant result, leaving out step markers and errors. Very long results are cut at about 24,000 characters.
• Every built-in workflow except Plan, build, verify reads it, so you can start a pipeline after a stage too.
• A new chat starts clean. Start one when the next job has nothing to do with the last.
• A single-agent answer counts as a result too, so a stage can build on a quick question you asked first.

What each step can do

Every step runs under a profile that sets its tools. The profiles fall into three kinds:

• Reads only. Research, Analyze, Plan, Architect, Security and Review. They read your workspace (and some search the web) but cannot write files or run commands.
• Runs checks, never edits. Debug and Verify cannot write files. They may run approved, non-destructive commands to reproduce a failure or check a result.
• Can change files. Build, Refactor, Optimize and Document.

While a chat follows a workflow, the skill menu stays on build and reads tools set per step. Each step's profile decides its tools, and a skill such as plan would otherwise limit what every step can use. Your own skill choice comes back when you switch to Single agent. Permission prompts still apply: commands ask before they run. See Permissions & safety.

A model for each step

With automatic model selection on, a workflow does not run on one model. Each step is rated on its own just before it starts, so Research and Verify can run on a light model while Architect runs on a heavy one, in the same run. See A model for each step.

Let Rusty choose

Choose Auto in the Mode menu and Rusty picks the workflow, or the single agent, for each message. It can also hand a running workflow over to another when what a step found changes the plan. Both are covered in Auto flows.

Make it yours

The built-in workflows are read-only, so they always stay in a known-good state. Choose Customize a copy in the Behaviors tab to get an editable version saved in your workspace, or build your own from scratch. Workflows and profiles are plain JSON. See Profiles & workflows.