Appearance
Steps & Workflows
Steps
A step is the reusable unit of work in Agentfleet: a natural-language instruction bound to exactly one tool.
| Field | Purpose |
|---|---|
| Name & description | What operators see in lists and run views. |
| Instruction | The model's contract: what to accomplish, what inputs to use, what to return. |
| Tool | The single tool the step may call. Its resource and scope determine the step's delegated token. |
Writing good instructions
The instruction is what the model reasons over — treat it like a specification:
- Name the inputs it should read from the run context (e.g. "the candidate FSP number from the run inputs").
- Define the success condition explicitly. If a tool returns a staged/pending outcome that is success (e.g.
StagedForReviewon a reviewed action), say so: "a StagedForReview result IS the successful outcome — do not retry." Without this, the model may treat it as a failure and retry. - Say what to return. Later steps see this step's outputs; tell the model what to surface.
One step, one tool
Steps are deliberately single-tool. This keeps the security boundary crisp — each step's token is scoped to exactly one operation surface — and keeps steps composable and reusable across workflows.
Workflows
A workflow is an ordered sequence of steps. Steps are reusable: the same step can appear in many workflows.
- Compose: create the workflow, then add steps with an explicit execution order. Reorder or remove at any time.
- Activate: only active workflows can be run. Deactivate to pause new runs without losing history.
- Data flows through context: each step's outputs are promoted into the run's context channel so subsequent steps can reference them — there is no hidden shared state.
Execution limits
Runs execute under deterministic guardrails (tool rounds per step, dynamic steps per task, a task timeout, an instance token budget, and context size caps). A run that breaches a limit fails explicitly with the limit recorded as the failure reason — it never silently truncates.
Related
- Runs — what happens when a workflow executes.
- Run Inputs — parameterizing a run.