openspec（OpenSpec 规范）

OpenSpec — Spec-Driven Development

OpenSpec structures AI-assisted development into trackable changes with artifacts (proposal, specs, design, tasks) that guide implementation.

Setup

# Install globally
npm install -g @fission-ai/openspec@latest

# Initialize in a project
cd /path/to/project
openspec init --tools claude

# Update after CLI upgrade
openspec update

Core Workflow

Each change follows: new → plan → apply → verify → archive

1. Start a Change

# Create change folder with default schema
openspec new change <name>

# With specific schema
openspec new change <name> --schema tdd-driven

2. Plan (Create Artifacts)

Use the CLI instructions command to get enriched prompts for each artifact:

# Get instructions for next artifact
openspec instructions --change <name> --json

# Check progress
openspec status --change <name> --json

Artifact sequence (spec-driven schema):

1. proposal.md — Why and what (intent, scope, approach)
2. specs/ — Requirements + scenarios (Given/When/Then)
3. design.md — Technical approach and architecture decisions
4. tasks.md — Implementation checklist with checkboxes

3. Implement

Read tasks.md and work through items, marking [x] as complete.

4. Verify

openspec validate --change <name> --json

Checks completeness, correctness, and coherence.

5. Archive

openspec archive <name> --yes

Merges delta specs into main openspec/specs/ and moves change to archive.

Agent Workflow (How to Use as an AI Agent)

When the user asks to build/migrate/refactor something with OpenSpec:

1. Check project state:

   openspec list --json           # Active changes
   openspec list --specs --json   # Current specs
   openspec schemas --json        # Available schemas

1. Create the change:

   openspec new change <name> [--schema <schema>]

1. For each artifact, get instructions and create the file:

   openspec instructions <artifact> --change <name> --json
   openspec status --change <name> --json

Then write the artifact file to openspec/changes/<name>/.

1. Implement tasks from tasks.md.

1. Validate and archive:

   openspec validate <name> --json
   openspec archive <name> --yes

CLI Quick Reference

Always use --json for programmatic/agent use.

Custom Schemas

Schemas define artifact sequences. Create custom ones for different workflows:

# Fork built-in schema
openspec schema fork spec-driven my-workflow

# Create from scratch
openspec schema init my-workflow

# Validate
openspec schema validate my-workflow

Schema files live in openspec/schemas/<name>/schema.yaml with templates in templates/.

For schema structure details, see references/schemas.md.

Project Structure

project/
├── openspec/
│   ├── config.yaml          # Project config (default schema, context, rules)
│   ├── specs/               # Source of truth — current system behavior
│   ├── changes/             # Active changes (one folder each)
│   │   └── <change-name>/
│   │       ├── .openspec.yaml
│   │       ├── proposal.md
│   │       ├── specs/       # Delta specs (what's changing)
│   │       ├── design.md
│   │       └── tasks.md
│   └── schemas/             # Custom schemas
└── .claude/skills/          # Auto-generated Claude integration

Spec Format

Specs use RFC 2119 keywords (SHALL/MUST/SHOULD/MAY) with Given/When/Then scenarios:

### Requirement: User Authentication
The system SHALL issue a JWT token upon successful login.

#### Scenario: Valid credentials
- GIVEN a user with valid credentials
- WHEN the user submits login form
- THEN a JWT token is returned

Delta Specs

Changes don't rewrite specs — they describe deltas (ADDED/MODIFIED/REMOVED) that merge into main specs on archive.

Config

openspec/config.yaml sets defaults:

schema: spec-driven      # or tdd-driven, rapid, custom
context: |
  Tech stack: TypeScript, React, Node.js
  Testing: Jest
rules:
  proposal:
    - Include rollback plan
  specs:
    - Use Given/When/Then format