gsd-pi: Plan and Run Coding Work with an AI Agent

Overview

GSD Pi coordinates AI-assisted software work from planning through implementation and verification. It addresses the problem of coding agents losing project context or making progress without a reviewable plan by organizing work into milestones, slices, and tasks, and recording project state locally.

It is aimed at developers who want an agent workflow integrated with their existing project and Git setup. Work happens through a terminal interface by default, with an optional web interface, provider integrations, and extension points for additional tools and commands.

Key Features

• Plan projects as milestones, slices, and tasks, then use auto mode to advance work.
• Run planned or quick coding sessions with project context and verification steps.
• Isolate implementation work in Git worktrees while keeping changes reviewable.
• Store requirements, decisions, plans, summaries, and validation evidence under .gsd/.
• Route requests to supported model providers, including API, OAuth, and external CLI providers.
• Extend workflows with commands, tools, skills, and UI integrations.
• Use a terminal interface or launch the optional web interface with gsd --web.

Use Cases

• A solo developer wants an agent to carry a feature from a written plan through implementation and validation.
• A team wants long-running coding tasks divided into reviewable units with local records of decisions and evidence.
• A developer wants to test an idea quickly using a one-off coding session rather than set up a full project plan.
• A team with existing model-provider preferences wants to use different provider integrations within a consistent workflow.

Project Facts

• Language: TypeScript
• License: MIT
• Stars: 1.3k
• Forks: 108
• Topics: context-engineering, meta-prompting, spec-driven-development
• Archived: No

Getting Started

Install and launch the guided setup:

npx @opengsd/gsd-pi@latest

For setup options, provider configuration, and development instructions, see the README.

Alternatives

• loopx: LoopX focuses on preserving goals, decisions, and handoffs across agent sessions, while gsd-pi organizes implementation into milestones and tasks.
• TaskSync: TaskSync centers on task queues and feedback loops, with VS Code and MCP options, rather than gsd-pi’s milestone-based CLI workflow.
• compound-engineering-plugin: Compound Engineering provides reusable planning, implementation, and review skills for multiple agent hosts; gsd-pi is a stateful CLI workflow.

Considerations

• GSD Pi is a workflow layer for AI coding, so its usefulness depends on configuring a supported model provider or external agent CLI.
• Project planning and runtime state are stored locally in .gsd/; the README also notes that migration snapshots in .gsd-backups/ are not removed automatically.
• The repository describes a new development baseline at version 1.0.0, with earlier release history archived separately.
• Worktree automation and autonomous progression add workflow structure, but users should review generated changes and verification results.