nicobailon/pi-subagents

What it solves

Pi-subagents allows a primary AI session (the parent) to delegate specific tasks to focused child agents. This prevents the main conversation from becoming cluttered and allows for specialized roles—such as code review, research, and implementation—to operate in parallel or in the background, providing a second set of "model eyes" on a project.

How it works

The system treats the main Pi session as a parent and spawns focused child sessions (subagents) with specific roles. When a user asks for a subagent, the parent Pi session decides which agent to use and how to compose the work. These subagents can run in the foreground (streaming progress directly into the conversation) or in the background (continuing to work while the user interacts with the main session).

Who it’s for

Developers and users of Pi who want to automate complex workflows, such as the "clarify $\rightarrow$ scout $\rightarrow$ worker $\rightarrow$ reviewer" loop, or those who need parallel audits and specialized AI assistance for codebase exploration and implementation.

Highlights

• Built-in Specialized Agents: Includes roles like scout (codebase recon), researcher (web/docs research), worker (implementation), reviewer (code review), and oracle (second opinions).
• Async Background Work: Ability to run tasks in the background and monitor them via a FleetView or a live inspector.
• Parallel Execution: Support for running multiple reviewers or agents simultaneously to check for different criteria (e.g., correctness, tests, and complexity).
• Orchestration Patterns: Provides prompt shortcuts and scripted workflows for repeatable patterns like review loops.
• Bounded Orchestration: Includes a maxSubagentSpawnsPerRun limit to prevent uncontrolled recursion.