English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

GSD (Get Shit Done): AI Coding Workflow Guide

Forum topic · 小凯 · 2026-03-12

Summary

GSD (Get Shit Done) is a spec-driven development framework for AI coding tools such as Claude Code, OpenCode, and Gemini CLI, with around 64K+ stars on GitHub. It addresses Context Rot by isolating each task in a fresh 200K token context window, so output quality does not degrade as conversations grow. The system includes 28+ slash commands, 12 specialized agents (planning, execution, verification, research), and a Node.js CLI that operates on plain Markdown files with no proprietary runtime. Its five-phase workflow covers initialization (/gsd:new-project), discussion (/gsd:discuss-phase N), planning (/gsd:plan-phase N), execution (/gsd:execute-phase N) with wave-based parallelism, and verification (/gsd:verify-work N). Generated artifacts include PROJECT.md, REQUIREMENTS.md, ROADMAP.md, PLAN.md, and STATE.md. Strengths include stable quality, atomic commits, parallel task execution, and no vendor lock-in, while limitations involve manual command triggers, a learning curve, and dependency on the Claude Code ecosystem.

GSD (Get Shit Done): AI Coding Workflow Guide

GSD (Get Shit Done) is one of the most popular AI coding workflows available today, with approximately 64K+ stars on GitHub. It is a Spec-Driven Development framework designed specifically for AI coding tools such as Claude Code, OpenCode, and Gemini CLI.

Core Problem Solved: Context Rot

  • In traditional AI programming, as a conversation progresses, the context window fills up and the AI's output quality gradually degrades.
  • GSD solves this with an "isolated context per task" approach: every task runs in a fresh 200K token environment.
  • Architecture

  • 28+ slash commands
  • 12 specialized agents (planning, execution, verification, research, etc.)
  • Plain Markdown files plus a Node.js CLI, no proprietary runtime
  • ---

    Five-Phase Core Workflow

    | Phase | Command | Purpose | |-----|------|-----| | Initialization | /gsd:new-project | Deep questioning → domain research → requirement extraction → roadmap generation | | Discussion | /gsd:discuss-phase N | Clarify implementation preferences, visual style, API design, etc. | | Planning | /gsd:plan-phase N | Research implementation approach, generate 2-3 atomic task plans | | Execution | /gsd:execute-phase N | Parallel plan execution, isolated context per task, atomic commits | | Verification | /gsd:verify-work N | Confirm objectives are met, not just tasks completed |

    Execution Modes

  • Wave-based parallelism: analyze dependencies, run independent tasks in parallel.
  • Quick mode /gsd:quick: skips the full flow for small tasks.
  • ---

    Common Commands

    | Command | Purpose | |------|------| | /gsd:new-project [--auto] | Create a new project; auto mode skips interaction | | /gsd:map-codebase | Analyze an existing codebase (legacy projects) | | /gsd:discuss-phase N | Discuss phase N implementation details | | /gsd:plan-phase N [--prd] | Create an execution plan for phase N | | /gsd:execute-phase N [--gaps-only] | Execute the phase N plan | | /gsd:verify-work N | Verify phase N work | | /gsd:quick [--discuss] [--full] | Quick mode for small tasks | | /gsd:complete-milestone | Archive, tag, and start the next cycle |

    ---

    Generated Core Documents

  • PROJECT.md - project vision
  • REQUIREMENTS.md - requirements with REQ-IDs
  • ROADMAP.md - phase breakdown and success criteria
  • PLAN.md - atomic task plans (XML structure)
  • STATE.md - project status tracking
  • ---

    Pros and Cons

    Strengths

    1. Solves Context Rot - each task gets an independent 200K context, keeping quality stable. 2. Atomic commits - every task is a separate commit, keeping git history clean. 3. Parallel execution - independent tasks run in parallel, improving efficiency. 4. Quality verification - every phase has a verification step to ensure objectives are met. 5. No vendor lock-in - pure Markdown plus standard Claude Code features. 6. Industry adoption - used by engineers at Amazon, Google, Shopify, Webflow, and others.

    Limitations

    1. Not fully automated - human-triggered commands are needed at each phase boundary. 2. Learning curve - requires understanding the spec-driven development mindset. 3. Heavy upfront investment - the initialization phase requires answering many questions. 4. Not suited for exploratory coding - better for projects with well-defined goals. 5. Claude Code dependency - the ecosystem is primarily built around Claude Code.

    ---

    One-Line Summary

    GSD is a framework that "turns chatting into engineering." It sacrifices some immediacy and conversational fluidity in exchange for a predictable, verifiable, and scalable AI coding experience. It is best suited for delivering production-grade code rather than rapid prototype validation.

    ---

    References

  • GitHub: https://github.com/gsd-build/get-shit-done
  • Install: npx get-shit-done-cc
  • Docs: https://zread.ai/gsd-build/get-shit-done

Tags

#gsd#ai-coding#claude-code#spec-driven-development#workflow#context-management#developer-tools#nodejs

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177168811