GSD hits 48K stars with spec-driven dev system for Claude Code

Context rot is one of those problems every developer hits with AI coding tools, and nobody talks about enough. As Claude fills its context window, output quality drops: responses get shorter, instructions get missed, and code gets sloppy. Get Shit Done (GSD) is the most serious open-source attempt I've seen at solving it structurally.

Built by developer TÂCHES, GSD adds structured context engineering and spec-driven workflows on top of Claude Code and now supports 12 runtimes in total. It just crossed 48.4K GitHub stars, which tells me this problem is more widespread than most teams admit.

What Happened

GSD v1.34.2 shipped on April 6, 2026, marking the project's 1,693rd commit across 47 releases since its initial commit in December 2025. That's a significant amount of development velocity for a project that's only a few months old.

The system installs via npx get-shit-done-cc@latest and works across Claude Code, OpenCode, Gemini CLI, Kilo, Codex, Copilot, Cursor, Windsurf, Antigravity, Augment, Trae, and Cline. With 119 contributors, 4K forks, and documentation in five languages, this has moved well past side-project territory.

What I find notable is how fast it got there. The growth curve from December 2025 to 48K stars by April 2026 reflects genuine utility, not just Twitter virality.

Key Features

• Multi-agent orchestration. Spawns parallel researchers, planners, executors, and verifiers, each in a fresh context window with up to 200K tokens dedicated to implementation. Your main session stays lean while subagents do the heavy lifting.
• Wave-based parallel execution. Groups independent plans into waves that run simultaneously, with dependent plans waiting in later waves and automatic commit-per-task tracking. This is the feature that makes large tasks actually manageable.
• Four canonical gate types (new in v1.34.0). Pre-flight, revision, escalation, and abort gates are wired into the plan-checker and verifier agents. Quality enforcement without manual review at every step.
• 12 runtime support. Single installer covers Claude Code, OpenCode, Gemini CLI, Kilo, Codex, Copilot, Cursor, Windsurf, Antigravity, Augment, Trae, and Cline. If your team uses more than one AI coding tool, you're covered.
• Built-in security pipeline. Prompt injection scanning, base64 payload detection, secret scanning, and a PreToolUse guard hook for .planning/ writes. The security layer is more thoughtful than most tools at this stage.
• SDK for headless execution. A TypeScript SDK (gsd-sdk) enables autonomous init and auto commands outside interactive sessions, which opens up CI/CD integration.

Why It Matters

Most AI coding tools treat context as infinite. It isn't. GSD structures work so each execution task gets a clean context window, which is a fundamentally different approach to reliability than just hoping the model holds it together.

The practical result is real: your main session stays at 30-40% context usage while subagents handle the work. Teams adopting spec-driven patterns get atomic git commits per task, traceable plans, and automated verification against stated goals. According to the README, the project is trusted by engineers at Amazon, Google, Shopify, and Webflow.

The broader pattern I'm seeing here is developers moving past raw prompting toward structured systems that manage context, enforce quality gates, and maintain state across sessions. GSD is one of the clearer expressions of that shift.

Example Use Case

A solo developer building a SaaS app with Next.js and Prisma runs /gsd-new-project to define their v1 scope. For Phase 1 (user authentication), they run /gsd-discuss-phase 1. GSD asks about password requirements, OAuth providers, and session handling.

Then /gsd-plan-phase 1 spawns researchers to investigate JWT libraries, creates 2-3 atomic plans with XML task structures, and verifies each plan against requirements. Running /gsd-execute-phase 1 groups independent plans into parallel waves: the user model and product model are built simultaneously, then dependent APIs follow. Each task gets its own git commit.

The developer walks away and comes back to working auth with a clean commit history. That outcome, repeatable and traceable, is what separates this from ad hoc prompting.

Competitive Context

GSD positions itself directly against other spec-driven tools. The README names BMAD and Speckit specifically, arguing that they add unnecessary enterprise ceremony: sprint ceremonies, story points and stakeholder syncs. GSD takes the opposite approach. Complexity lives in the system, not the workflow.

What I find more interesting is what it adds on top of Claude Code itself. Claude Code alone doesn't manage context rotation, enforce quality gates, or maintain planning state across sessions. GSD fills that gap without requiring an entirely different tool. The inherit model profile option also lets teams use non-Anthropic providers through OpenRouter or local models, which decouples the workflow from any single LLM vendor.

Vendor flexibility is increasingly important as teams hedge across providers. GSD being model-agnostic by design is a deliberate and smart call.

My Take

GSD is a context-engineering system that makes AI coding sessions more reliable by managing what the model sees and when it sees it. If you use Claude Code or any of the 12 supported runtimes for serious development work and have hit context degradation on larger tasks, this is worth evaluating.

Install with npx get-shit-done-cc@latest and run /gsd-help to start. At 48.4K stars and 119 contributors, the community weight behind it is real. I'd rather adopt a system that's already been stress-tested than build the same scaffolding myself.