OpenSpec
Overview
OpenSpec is an open-source, lightweight framework for spec-driven development (SDD) with AI coding assistants, developed by Fission AI. Rather than keeping requirements in ephemeral chat history, OpenSpec introduces a structured specification layer so humans and AI align on requirements before implementation begins. The framework is designed to be fluid and iterative — not rigid or waterfall — and serves both brownfield and greenfield projects at any scale. As of April 2026, the project had 50.1k GitHub stars and is actively maintained under the MIT license.
(Updated: the page previously described OpenSpec as "upcoming" and in "early development"; as of v1.3.1 it is a production-grade tool with a large community.)
Core Design Philosophy
Three foundational principles guide OpenSpec:
- Agree before building — specifications are written and reviewed before implementation starts, reducing back-and-forth corrections.
- Stay organized — every change proposal lives in a dedicated folder with consistent artifact structure, keeping context outside chat history.
- Work fluidly — any artifact can be updated at any time; there are no rigid phase gates between specification and implementation.
The philosophy is summarized by its maintainers as: "fluid not rigid, iterative not waterfall, easy not complex, built for brownfield not just greenfield, scalable from personal projects to enterprises."
Artifact Structure
Each change proposal creates a dedicated directory containing four artifact types:
| Artifact | File | Purpose |
|---|---|---|
| Proposal | proposal.md |
Rationale, scope, and overview of the change |
| Specifications | specs/ directory |
Requirements and scenario-driven acceptance criteria |
| Design | design.md |
Technical approach and architectural decisions |
| Tasks | tasks.md |
Numbered implementation checklist for the AI agent |
Completed proposals are moved to a timestamped archive folder via /opsx:archive, keeping the working directory clean.
Primary Workflow: /opsx Commands
OpenSpec uses slash commands as its primary interface for artifact-guided development:
| Command | Action |
|---|---|
/opsx:propose <idea> |
Generates the full change folder (proposal, specs, design, tasks) |
/opsx:apply |
Executes all tasks systematically — component creation, styling, integration |
/opsx:archive |
Moves completed change to archive; updates specs for future iterations |
/opsx:new |
Starts a new change proposal |
/opsx:continue |
Resumes work on an existing proposal |
/opsx:ff |
Fast-forwards through multiple tasks |
/opsx:verify |
Validates implementation against the specification |
/opsx:bulk-archive |
Archives multiple completed changes at once |
/opsx:onboard |
Onboards an existing project into OpenSpec |
Installation and Requirements
npm install -g @fission-ai/openspec@latest
cd your-project
openspec init
- Node.js: 20.19.0 or higher required
- Package managers: npm, pnpm, yarn, bun, and nix supported
- Language: TypeScript (99.1% of codebase)
- Latest version: v1.3.1 (April 21, 2026)
- Update:
openspec updateregenerates agent instructions and activates latest slash commands
Telemetry: Anonymous telemetry collects only command names and version numbers; automatically disabled in CI environments; opt out via OPENSPEC_TELEMETRY=0.
Tool Integration
OpenSpec works with 25+ AI assistants and development tools, including Claude, GitHub Copilot, VS Code, Cursor, and others. This tool-agnosticism is a deliberate design choice — it does not lock users into a specific IDE or model vendor.
Community schemas extend the base framework: third-party opinionated workflow bundles are distributed as standalone repositories, allowing teams to share and version their own conventions on top of OpenSpec primitives.
Comparison with Alternatives
| Dimension | OpenSpec | GitHub Spec Kit | AWS Kiro |
|---|---|---|---|
| Weight | Lightweight, minimal setup | Heavier, Python dependency | Proprietary IDE-centric |
| Phase gates | None — fluid updates throughout | Rigid phase structure | Structured to Kiro workflows |
| Tool scope | 25+ AI assistants, tool-agnostic | GitHub-native | Kiro IDE / AWS toolchain |
| Project type | Brownfield + greenfield | Primarily greenfield | AWS-integrated projects |
| Open source | MIT | N/A | Proprietary |
| Installation | npm install -g |
Python setup | IDE plugin |
OpenSpec's primary differentiator is that it solves unpredictability when requirements live only in chat history, without imposing a heavyweight process. The Thoughtworks Technology Radar (Vol. 34) categorizes it as "New" and notes it "focuses on spec deltas rather than complete upfront specification — well-suited for brownfield systems."
Best Practices
| Challenge / Area | Description | Solution / Recommendation |
|---|---|---|
| Model selection | Not all models handle spec generation equally | Use high-reasoning models; Claude Opus 4.7 and Codex 5.5 recommended |
| Context management | Long chat history degrades AI spec quality | Start with clean context windows before /opsx:apply |
| Architectural changes | Ad-hoc AI edits to architecture cause drift | Submit an OpenSpec proposal before implementing any architectural change |
| Brownfield adoption | Existing projects lack a specification baseline | Use /opsx:onboard to generate initial specs from the current codebase |
| Spec hygiene | Artifacts become stale if not maintained | Archive completed proposals; update specs on each iteration before re-proposing |
Use Cases
- Feature development: Propose, specify, and implement discrete features with full artifact trail
- Brownfield refactoring: Onboard existing codebases and incrementally capture specs for change areas
- API design: Write specs for endpoints and let AI generate implementation against them
- Multi-session work: Artifacts persist across sessions, giving AI agents recoverable context
- Team collaboration: Shared spec folder in version control provides single source of truth for requirements
See Also
- Agentic AI Foundation
- AGENTS.md Standard
- Model Context Protocol
- Agent2Agent (A2A) Protocol
- AG-UI Protocol
- AIDLC Workflows (AWS)
- AgentHarness Engineering
- Context Engineering
- Production Best Practices: Testing & Evaluations
References
- Fission-AI/OpenSpec GitHub Repository — official source, MIT license, v1.3.1
- Thoughtworks Technology Radar Vol. 34 — categorized as "New"; assessment of OpenSpec vs. BMAD and Kiro