PromptPlan Workflows Explained: Inputs, Steps, Runs, and Outputs

What a workflow contains

A workflow is the main reusable work object in PromptPlan. Each workflow contains:

• metadata such as title, summary, category, visibility, and status
• workflow inputs
• ordered steps
• model policy per step
• saved runs
• final outputs and step-by-step outputs

Workflow inputs

Inputs are the structured fields a user fills before running the workflow.

Supported input types in the MVP:

• text
• long_text
• number
• boolean
• select
• json

Each input can include:

• key
• label
• required flag
• default value
• help text
• placeholder
• sort order

Workflow steps

Steps run in order and form the execution chain.

Supported step types:

• input
• transform
• reason
• generate
• review
• export

Each step can define:

• key
• title
• description
• prompt template
• output format
• provider
• model name
• temperature
• max tokens
• enabled or disabled state

Workflow runs

When a workflow runs, PromptPlan creates:

• a WorkflowRun
• a saved snapshot of the workflow definition
• a saved snapshot of run inputs
• a WorkflowStepRun record for each executed step

This matters because old runs stay readable even after the workflow is edited later.

Final output rule

PromptPlan chooses the final output using this order:

1. the last successful export step
2. otherwise the last successful generate or review step
3. otherwise the last successful completed step

Visibility and ownership

Workflows are user-owned.

Current workflow visibility values:

• private: owner-only
• unlisted: accessible by direct link for signed-in users when not archived
• public: model exists in the schema, but public discovery is not yet the main surface

What the current MVP does not do

The current workflow system is intentionally constrained:

• linear execution only
• no branching
• no loops
• no autonomous hidden planning layer
• no local/Ollama private execution yet

Those constraints make runs easier to understand, debug, and improve.