Superpowers, OpenSpec, and Spec‑Kit: How to Distinguish and Choose the Right AI Coding Tool

Problem: Vibe Coding

AI coding agents often start implementing code without fully understanding requirements, skipping design, testing, or code review. This “Vibe Coding” stems from a lack of structured workflow constraints.

Three Tools that Impose Discipline

Spec‑Kit – defines what rules to follow (executable specifications).

OpenSpec – tracks what changed (lightweight spec management).

Superpowers – governs how to do it (skill‑driven workflow).

Superpowers

What It Is

Superpowers is an AI‑coding agent skill framework and methodology (creator: Jesse Vincent, GitHub: obra). It does not make the model smarter; it forces the AI to follow disciplined steps similar to a project manager: think, plan, code, review.

GitHub stats (June 2026): 238 K stars, 21.1 K forks; Anthropic plugin marketplace shows 680 K installations. Supports Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot CLI, etc.

Core Mechanism – 14 Composable Skills

brainstorming – entry skill; AI asks one clarification question at a time, proposes 2‑3 designs, and produces a design document only after explicit user approval.

subagent‑driven‑development – spawns an isolated sub‑agent per task and applies a two‑stage review (spec compliance + code‑quality). Failed reviews send the task back for redo.

test‑driven‑development – enforces the red‑green‑refactor cycle.

requesting‑code‑review – automatically fetches the current git diff, reviews each file, and blocks progress on severe issues.

Five‑Stage Development Process

Every code output must pass the following stages, none may be skipped:

Brainstorming (design clarification).

Planning (task breakdown).

Implementation (sub‑agent execution).

Testing (TDD cycle).

Code review (automated diff analysis).

Rule: No code is considered complete until all five steps are finished.

Pros & Cons

Pros: enforces engineering discipline; 14 skills cover the full lifecycle; sub‑agent isolation prevents context pollution; two‑stage review ensures quality; platform‑agnostic (Markdown).

Cons: rigid skills may limit AI creativity; mandatory TDD adds initial overhead; learning curve for 14 skills.

OpenSpec

What It Is

OpenSpec is an AI‑native spec‑driven development framework created by the Fission AI team. It makes AI coding tools follow a structured spec document rather than generating code arbitrarily.

GitHub stats (June 2026): 56 K+ stars; supports Node.js 20.19.0+. Tool‑agnostic, suited for brown‑field projects.

Core Mechanism – Incremental Specs

OpenSpec introduces delta‑based specs with four change types:

ADDED – new behavior appended.

MODIFIED – replace an existing requirement block.

REMOVED – delete a requirement block.

RENAMED – rename using FROM:/TO: syntax.

This design is especially friendly to existing codebases where requirements evolve incrementally.

Four Core Skills

/opsx:propose

– create a change proposal, generate design/specs/tasks. /opsx:explore – exploratory mode to clarify requirements and design ideas. /opsx:apply – implement tasks step‑by‑step from tasks.md. /opsx:archive – archive completed change and merge the spec.

Full Workflow Example (Weather CLI)

Step 1: Propose change

/opsx:propose "Create a simple weather‑query CLI tool"

OpenSpec generates four files: proposal.md – Why/What/Impact. specs/city-weather-query/spec.md – detailed functional requirements. design.md – architecture decisions (uses wttr.in API, no external deps). tasks.md – executable checklist.

Step 2: Human reviews the markdown files.

Step 3: Execute – run /opsx:apply to let the AI implement tasks.

Step 4: Archive – run /opsx:archive to merge and close the change.

DAG Dependency Graph

OpenSpec internally uses a directed acyclic graph to enforce the order: proposal → spec → design → tasks. The DAG prevents out‑of‑order execution.

Pros & Cons

Pros: lightweight; incremental specs; traceable change lifecycle; tool‑agnostic.

Cons: execution layer is weaker (focuses on WHAT, not detailed HOW); no mandatory quality gates; may leave implementation details to the AI.

Spec‑Kit

What It Is

Spec‑Kit is an official GitHub‑backed spec‑driven development toolkit.

GitHub stats (June 2026): 115 K+ stars, 10.2 K forks; supports 25+ AI agents (Claude Code, Copilot, Cursor, Gemini, etc.).

Core Mechanism – Seven‑Stage Pipeline

The pipeline defines explicit inputs and outputs for each stage, similar to a factory line:

constitution.md – project‑level governance (code quality, testing standards, UX consistency, performance).

spec.md – functional requirements (what and why, no tech stack).

plan.md – technical solution (stack choice, architecture, API contracts).

tasks.md – executable task list derived from the plan.

Installation & Usage

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install Specify CLI
uv tool install specify-cli --from

Slash commands inside an AI agent trigger each stage:

/speckit.constitution   # define project constitution
/speckit.specify        # write functional spec
/speckit.plan           # create technical plan
/speckit.tasks          # break down tasks
/speckit.analyze        # consistency checks
/speckit.implement      # execute implementation

Pros & Cons

Pros: GitHub backing; executable specs; clear seven‑stage pipeline; supports 25+ AI agents; project‑level constitution.

Cons: steep learning curve; geared toward green‑field projects; Python‑centric (may be a barrier for Java teams).

Side‑by‑Side Comparison (Key Dimensions)

GitHub Stars : Superpowers ≈ 238 K, OpenSpec ≈ 56 K, Spec‑Kit ≈ 115 K.

Core Focus : Superpowers – skill‑driven workflow; OpenSpec – lightweight spec management; Spec‑Kit – executable specifications.

Problem Solved : Superpowers – “how to do it”; OpenSpec – “what changed”; Spec‑Kit – “what rules to follow”.

Core Mechanism : Superpowers – 14 skills + five‑stage process; OpenSpec – delta‑based specs + DAG; Spec‑Kit – seven‑stage pipeline.

Tech Stack : Superpowers – Markdown (cross‑platform); OpenSpec – TypeScript/Node.js; Spec‑Kit – Python + uv.

Best For : Superpowers – all project types; OpenSpec – brown‑field (existing codebases); Spec‑Kit – green‑field (new projects).

Learning Curve : Superpowers – medium; OpenSpec – low; Spec‑Kit – steep.

AI Tool Support : Superpowers – Claude Code, Cursor, Codex, Gemini, etc.; OpenSpec – Cursor, Copilot, Claude Code, etc.; Spec‑Kit – 25+ agents.

Official Backing : Superpowers – Anthropic plugin marketplace; OpenSpec – community driven; Spec‑Kit – GitHub official.

When to Choose Each Tool

Superpowers

Project demands strict engineering discipline (forced TDD, code review, sub‑agent isolation).

High complexity with many files and parallel tasks.

Desire for AI to act like a senior engineer rather than a simple code generator.

Tool‑agnostic across multiple AI backends.

OpenSpec

Existing codebase (brown‑field) that needs incremental change tracking.

Requirement traceability is critical (proposal → spec → design → tasks → archive).

Team collaboration with clear change documentation.

Lightweight setup; no API‑key requirement.

Spec‑Kit

Starting a brand‑new project (green‑field) where a full‑stack pipeline is valuable.

Need for GitHub’s official backing and long‑term maintenance.

Desire for specifications that directly generate working code.

Team willing to invest time learning a seven‑stage process.

Combining the Tools

A practical “three‑in‑one” workflow:

Use Spec‑Kit to define the project constitution (global rules).

For each new feature, run OpenSpec propose to create a change proposal.

Employ Superpowers brainstorming to clarify requirements.

Execute development with Superpowers subagent‑driven‑development.

Archive the change with OpenSpec archive.

This approach provides global governance (Spec‑Kit), change traceability (OpenSpec), and disciplined execution (Superpowers).

Open‑Source Repositories

Superpowers – https://github.com/obra/superpowers (238 K stars)

OpenSpec – https://github.com/Fission-AI/OpenSpec (56 K+ stars)

Spec‑Kit – https://github.com/github/spec-kit (115 K+ stars)