spec-compare — Summary

spec-compare is a research and comparison documentation repository, not a deployable AI coding framework. It contains in-depth analysis of spec-driven development tools for AI-assisted coding, organized as 13 Markdown documents covering comparison matrices, 12-scenario use-case scoring, git worktree analysis, a landscape survey of 30+ tools, and practical decision frameworks. The original six tools compared are Spec-Kit, Spec Kitty, BMad Method, OpenSpec, Kiro, and Tessl; a February 2026 extension adds GSD, Ralph Loop, Zencoder, Kilo Code, Conductor, and PromptX. The repository's primary original contribution is the "SDD Maturity Levels" taxonomy — Spec-First (specs precede code but are discarded), Spec-Anchored (specs persist and evolve), and Spec-as-Source (only specs are edited, code auto-generates) — which provides a conceptual framework for categorizing all tools in the SDD space. With 39 stars and 2 contributors, spec-compare has more adoption signal than most deployable frameworks in this batch, reflecting its utility as a curated research entry point for practitioners choosing between SDD tools.

differs_from_seeds

spec-compare has no relationship to any seed framework — it does not extend, fork, or build on any of the five archetypes. It is a pure methodology documentation artifact: no commands, no skills, no agents, no MCP servers, no hooks, and no install script. Its value is analytical rather than operational: it tells you which seed-class framework to choose, rather than being a framework itself. In this batch it is the only Tier C item — it does not fit the "AI coding agent framework" definition, but it is included because it is the most comprehensive publicly available comparison of the exact tool category this research project covers.

spec-compare — Overview

Origin

spec-compare was created by cameronsjo with a second contributor, last updated 2026-02-25. It is MIT-licensed with 39 stars. The repository has no source code — it is entirely documentation.

Purpose

From the README:

A comprehensive research and comparison of spec-driven development (SDD) tools for AI-assisted coding, including analysis of git worktree support, architectural approaches, and practical recommendations.

The repository was originally created to compare six tools (Spec-Kit, Spec Kitty, BMad Method, OpenSpec, Kiro, Tessl) and was extended in February 2026 to add GSD, Ralph Loop, Zencoder, Kilo Code, Conductor, and PromptX.

Primary Original Contribution

The "SDD Maturity Levels" taxonomy, from the README:

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)

Key Findings

From the README:

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 (but closed beta)
• Spec-Kit — Requires /speckit.clarify workaround, not optimized for small changes
• Kiro/BMad — "Sledgehammer to crack a nut" problem for trivial changes

Spec Kitty is the only tool with built-in git worktree support, enabling:

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

Key finding: OpenSpec v1.0 is the only tool that migrated from AGENTS.md/CLAUDE.md to the newer SKILL.md standard. All other open-source tools still generate AGENTS.md. The AGENTS.md standard itself (28.64% runtime reduction in evaluations) continues to gain adoption — OpenAI Codex ships 88 AGENTS.md files in its own repo.

spec-compare — Architecture

Distribution Type

Methodology documentation (static Markdown files). No deployable artifact.

Repository Structure

spec-compare/
├── README.md                        # Entry point with key findings, quick comparison table, recommendations
├── CHANGELOG.md                     # Version history of the research
├── CONTRIBUTING.md                  # How to contribute findings
├── LICENSE                          # MIT
└── docs/
    ├── comparison.md                # Side-by-side feature comparison matrices
    ├── use-case-scoring.md          # 12-scenario graded scoring (5-star scale)
    ├── iterative-development.md     # Spec modification workflows analysis
    ├── git-worktree-support.md      # Detailed worktree analysis (updated with Beads, Conductor)
    ├── recommendations.md           # Decision frameworks by use case
    ├── critical-analysis.md         # Concerns, critiques, future outlook
    ├── landscape.md                 # 30+ multi-agent tools surveyed
    ├── beads.md                     # Agent memory, messaging, multi-agent villages
    ├── gaps.md                      # Zencoder, Kilo Code, Conductor, PromptX analyses
    ├── cheatsheet-beads-openspec.md # Practical Beads + OpenSpec setup cheatsheet
    ├── sources.md                   # All citations and references
    ├── research-beads-agentmail-for-skill-update.md  # Research notes
    └── tools/
        ├── spec-kit.md              # Individual profile
        ├── spec-kitty.md            # Individual profile
        ├── bmad-method.md           # Individual profile
        ├── openspec.md              # Individual profile
        ├── kiro.md                  # Individual profile
        ├── tessl.md                 # Individual profile
        ├── gsd.md                   # Individual profile
        └── ralph-loop.md            # Individual profile

No Install, No Runtime, No Configuration

spec-compare has no install method, no runtime dependencies, and no configuration files. It is read directly on GitHub or cloned for offline reference.

spec-compare — Components

spec-compare contains no executable components. All content is static Markdown research documentation.

Document Index

Tools Analyzed (12 total)

Original Six

1. GitHub Spec-Kit — Open-source CLI toolkit for greenfield projects
2. Spec Kitty — Community fork with built-in git worktree orchestration
3. BMad Method — Enterprise framework with 21 specialized AI agents
4. OpenSpec — Lightweight change-management for brownfield projects
5. Kiro — AWS-backed agentic IDE with multimodal input
6. Tessl — Experimental spec-as-source platform

February 2026 Extensions

7. GSD — Meta-prompting SDD system with wave-based context management (11.9K stars)
8. Ralph Loop — Stateless iterative execution pattern by Geoffrey Huntley
9. Zencoder/Zenflow — Commercial SDD-as-a-Service platform
10. Kilo Code — Open-source agentic platform with Memory Bank ($8M seed, 1.5M users)
11. Conductor — macOS parallel agent runner using git worktrees
12. PromptX — AI agent context platform via MCP (gap entry)

spec-compare — Uniqueness and Positioning

differs_from_seeds

spec-compare has no relationship to any of the five seed archetypes — it does not extend, fork, or build on any seed framework. It is a pure methodology documentation artifact: a curated research collection designed to help practitioners choose between seed-class frameworks. No other repository in this batch serves this function. It is the only Tier C item in batch-01: included for completeness (and because its 39 stars reflect genuine practitioner adoption as a reference resource) but not analyzable as a deployable framework.

Positioning

• Category: Research/comparison documentation — NOT a deployable AI coding framework
• Primary value: Curated cross-tool analysis with graded use-case scoring and the "SDD Maturity Levels" taxonomy
• Unique contribution: The 3-level SDD Maturity taxonomy (Spec-First / Spec-Anchored / Spec-as-Source) provides conceptual vocabulary for categorizing every other framework in this research project
• Scope: Covers the widest span of any single document in the corpus — original 6 tools + 6 extended tools + 30+ landscape entries

Observable Limitations

1. Not a framework: Cannot be installed, configured, or run. Zero operational value for practitioners who already know which tool they want.
2. Point-in-time accuracy: Some entries (Kiro, Tessl) are based on preview/beta state as of early 2026 and may be stale.
3. Author bias unknown: Tool scoring reflects the author's priorities (brownfield/modification workflows ranked highly; this favors OpenSpec). Readers with greenfield priorities may weight the scoring differently.
4. No code evidence: The comparison relies on documentation claims rather than reproducible benchmarks. The "30+ minute workflow for button color" for BMad is an estimate, not a measurement.
5. Missing frameworks: Doesn't cover ralphy-openspec, goopspec, or the wave of OpenCode-targeted tools that emerged after February 2026.