Skip to content

Latest commit

 

History

59 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Spec-Driven Development Tools Comparison

Independent research comparing 19 spec-driven development (SDD) tools for AI-assisted coding — 8 with full workflow treatment, 11 emerging — plus the surrounding agent-orchestration and execution layer. Feature matrices, use-case scoring, git-worktree analysis, decision frameworks, and an interactive comparison site, all derived from the research docs in this repo.

A considered opinion, not a benchmark. This project is independent and unaffiliated with every vendor tracked; where a vendor initiated contact (Memex), the review says so up front.

🔮 Live site

An interactive, Artificer-themed visualizer of this research: cameronsjo.github.io/spec-compare

  • Compare workflows — step the same scenario (trivial change, greenfield, refactor, bug fix, parallel dev, cross-cutting) through all eight core tools in lockstep and watch OpenSpec's short path contrast with BMad's long one.
  • Feature matrix — sortable, filterable capability table aggregated live from the data.
  • Scoring heatmap — all nineteen tools across seven use-case dimensions, color-graded 1–5.
  • Decision guide — an interactive flowchart that walks you to a recommended tool.
  • Tool profiles — per-tool metadata, key features, and limitations.

Every value is extracted from the research docs below — no fabricated attributes.

Developing the site

The site lives in site/ — a React + Vite + TypeScript app on a vendored copy of the Artificer design system. Each tool is one JSON file under site/src/data/tools/, auto-discovered and AJV-validated against schema.json; the matrix and heatmap are derived by aggregating across those files (single source of truth).

cd site
npm install
npm run dev        # local dev server
npm run validate   # AJV-validate every tool JSON + cross-tool invariants
npm run build      # validate → tsc → vite build (outputs site/dist/)
npm run preview    # serve the production build locally

Pushing to main builds and publishes site/dist/ to GitHub Pages via .github/workflows/deploy.yml.

The tracked set

Core tools get the full treatment — workflow graphs, per-scenario walkthroughs, heatmap scores. Emerging tools get a profile card and heatmap row only. When each entered (and every re-verification since) is dated in the CHANGELOG and the reassessment docs.

Core (8)

  • GitHub Spec-Kit - Open-source CLI toolkit for greenfield projects
  • Spec Kitty - Community fork with built-in git worktree orchestration
  • BMad Method - Enterprise framework with 21 specialized AI agents
  • OpenSpec - Lightweight change-management for brownfield projects
  • Kiro - AWS-backed agentic IDE, GA since Nov 2025 (paid tiers + CLI)
  • Tessl - Spec-as-source platform; public Framework + Registry ($125M raised)
  • Superpowers - MIT skills framework + methodology; brainstorm → plan → subagent TDD (~214K stars)
  • Traycer - Commercial VS Code Plan → Execute → Verify layer over your agent (100K+ users)

Emerging (11)

  • GSD - Meta-prompting SDD system with wave-based context management (63.8K stars)
  • Ralph Loop - Stateless iterative execution pattern by Geoffrey Huntley
  • Zencoder/Zenflow - Commercial SDD control plane; free Zenflow desktop app
  • Kilo Code - Open-source agentic platform with Memory Bank ($8M seed, 1.5M users)
  • Conductor - macOS parallel agent runner using git worktrees
  • MoAI-ADK - Go CLI wrapping Claude Code in a SPEC-First Plan→Run→Sync lifecycle with TDD gates (~1.1K stars)
  • Frame - Electron "Agentic Development Environment" orchestrating parallel agents in isolated worktrees
  • GRACE - Contract-first Graph-RAG methodology as installable agent skills (XML artifacts, drift detection)
  • GAAI - Governed autonomous delivery: Discovery → git-tracked backlog → Delivery daemon (source-available, ELv2)
  • Smart Ralph - Claude Code/Codex plugin layering spec phases on the Ralph autonomous loop
  • Memex - Fair-code Specify & Verify platform: specs as a typed decision database with CI-verified acceptance criteria and a knowledge graph (Mindset AI; solicited review, see disclosure)

Two gap entries are tracked in prose but not scored: PromptX (AI agent context platform via MCP) and MUSUBI (maximally-rigorous SDD framework, ~57 stars, stalled) — see docs/gaps.md.

Key Findings

The Modification Problem

Critical Gap: Most SDD tools excel when requirements are clear upfront but struggle with iterative changes like "change button from blue to green."

  • OpenSpec - Purpose-built for modifications with delta format (ADDED, MODIFIED, REMOVED)
  • Tessl - Spec-as-source enables edit-and-regenerate (Framework now public)
  • Spec-Kit - Requires /speckit.clarify workaround, not optimized for small changes
  • Kiro/BMad - "Sledgehammer to crack a nut" problem for trivial changes

See Iterative Development Analysis and Use Case Scoring for details.

The Enforcement Gap

Agents ignore specs. Writing the spec is the easy half; most tools have nothing that makes the implementation honor it. The newest entrants all attack this from different angles — Traycer's Verify phase, MoAI-ADK's TDD gates, GRACE's drift detection, and Memex's per-criterion CI-verified acceptance criteria (the most direct design, and the least proven). See Critical Analysis and the Memex review.

Git Worktree Support

Spec Kitty pioneered built-in git worktree support among SDD tools (Superpowers, Conductor, Zencoder/Zenflow, MoAI-ADK, and Frame now automate worktrees too), enabling:

  • Automatic worktree creation per feature
  • Parallel feature isolation without branch switching
  • Automated cleanup on merge

SDD Maturity Levels

  1. Spec-First: Specs precede coding but are discarded (Spec-Kit, Kiro, BMad)
  2. Spec-Anchored: Specs persist and evolve (OpenSpec, Spec Kitty)
  3. Spec-as-Source: Only specs are edited, code auto-generates (Tessl)

Documentation

The research is organized into focused, digestible documents:

Individual Tool Profiles

  • GitHub Spec-Kit - Open-source CLI toolkit
  • Spec Kitty - Community fork with worktree support
  • BMad Method - Enterprise framework with 21 agents
  • OpenSpec - Lightweight change management
  • Kiro - AWS-backed agentic IDE (GA)
  • Tessl - Spec-as-source platform (public Framework + Registry)
  • Superpowers - Auto-triggered skills + methodology, any harness
  • Traycer - Spec-driven Plan → Execute → Verify layer (VS Code)
  • GSD (Get Shit Done) - Context-engineered SDD for solo developers
  • Ralph Loop - Stateless iterative execution pattern

Analysis & Recommendations

Orchestration & Execution Layer

Quick Comparison (core tools)

The full 19-tool version of this table is generated from the tool data in docs/comparison.md.

Tool License Git Worktrees Best For Maturity
Spec-Kit Open Source No Greenfield projects Production (v1.0.1)
Spec Kitty Open Source Yes Parallel development Active Dev (v3.2.5)
BMad Method Open Source No Enterprise workflows Stable (v6.11.0)
OpenSpec MIT No Brownfield changes Production (v1.11.0)
Kiro Proprietary No IDE experience GA (v1.0.x)
Tessl Proprietary No Spec-as-source Active Dev (public)
Superpowers MIT Yes Disciplined autonomous dev Active Dev (v6.3.0)
Traycer Proprietary No Plan-first orchestration Active Dev

Versions current as of 2026-08-28. The August sweep resolved the drift the July sweep had deferred — Kiro, Spec-Kit, and Superpowers are all bumped in place, with their workflow shapes re-confirmed rather than assumed — see the sweep doc.

Recommendations

For Git Worktree Users

Use Spec Kitty - The most complete built-in worktree management + parallel feature isolation among SDD tools (Superpowers also automates worktrees via its skill).

For Simplicity

Use OpenSpec - Lightweight change management without excessive overhead.

For Enterprise

Use BMad Method - Comprehensive workflows with 21 specialized agents.

For Greenfield Projects

Use Spec-Kit - Battle-tested, constitution-driven development.

For Experimentation

Try Kiro or Tessl - Kiro is GA (free tier + paid); Tessl's Framework + Registry is now public.

Critical Perspectives

The research includes analysis of:

  • The Waterfall Question: Does SDD reintroduce waterfall bureaucracy?
  • AI Adherence Issues: Agents frequently ignore specifications
  • Scalability Concerns: Unclear when SDD adds value vs. overhead
  • Historical Parallels: Similarities to failed Model-Driven Development (MDD)

Market Context

The early-2025 signals that framed this research, kept for provenance:

  • 25% of Y Combinator's Winter 2025 cohort reported 95% AI-generated codebases
  • Industry leaders predicted developers "won't look at code" by 2027
  • Specifications pitched as "the fundamental unit of programming"

Contributing

See CONTRIBUTING.md for contribution guidelines.

Changelog

See CHANGELOG.md for version history and changes.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Sources

All research is compiled from publicly available sources including:

  • Official tool documentation and repositories
  • Industry blog posts and articles
  • Comparative analyses from Martin Fowler, Medium, and others
  • Critical perspectives from Marmelab, RedMonk, and Thoughtworks

Full source citations are available in docs/sources.md.

Contact

For questions, issues, or suggestions, please open an issue on GitHub.


Last Updated: 2026-08-28

About

Interactive comparison of the spec-driven development tools — lockstep workflow walkthroughs, a scoring heatmap across use cases, and a decision guide, live at cameronsjo.github.io/spec-compare

Resources

Contributing

Stars

140 stars

Watchers

3 watching

Forks

Releases

Packages

Used by

Contributors

Languages