AgwaB/pi-workflow

What it solves

pi-workflow provides a way to run complex, multi-step AI tasks that are too large for a single prompt. It allows users to define repeatable, structured processes—such as deep research, code reviews, or specification checks—that coordinate multiple AI agents to ensure higher quality and more consistent results than a one-off interaction.

How it works

The system treats a workflow as a deterministic stage graph. It coordinates "subagent" workers across different stages, passing the results (artifacts) from one step to the next.

It supports several execution patterns:

• Single: A one-to-one prompt and agent interaction.
• Foreach: A dynamic "fan-out" where a list of items is generated and processed by multiple agents in parallel.
• Reduce: A "fan-in" where multiple results are synthesized into a final report.
• Loop: Bounded repetition until a specific condition is met.
• DAG: Nested graphs for complex organizational structures.
• Dynamic: Adaptive orchestration where controller code creates tasks on the fly.

Who it’s for

It is designed for developers and teams using Pi who need to automate rigorous technical routines like architecture trade-off analysis, API conformance checks, and risk assessments for proposed changes.

Highlights

• Bundled Workflows: Includes pre-made templates for deep-research, deep-review, spec-review, and impact-review.
• Execution Profiles: Allows users to save and switch between different model setups (e.g., Claude, Codex) for specific workflow stages.
• Workflow Board: A dedicated TUI (Terminal User Interface) to inspect run progress, drill into individual tasks, and view artifacts.
• Authoring Tools: Includes a workflow-guide skill to help users create and customize their own project-specific workflows using natural language.