Tabish Bidiwale
4.3 KiB
OpenSpec Conventions Specification
Purpose
OpenSpec conventions SHALL define how system capabilities are documented, how changes are proposed and tracked, and how specifications evolve over time. This meta-specification serves as the source of truth for OpenSpec's own conventions.
Core Principles
The system SHALL follow these principles:
- Specs reflect what IS currently built and deployed
- Changes contain proposals for what SHOULD be changed
- AI drives the documentation process
- Specs are living documentation kept in sync with deployed code
Directory Structure
WHEN an OpenSpec project is initialized THEN it SHALL have this structure:
openspec/
├── project.md # Project-specific context
├── README.md # AI assistant instructions
├── specs/ # Current deployed capabilities
│ └── [capability]/ # Single, focused capability
│ ├── spec.md # WHAT and WHY
│ └── design.md # HOW (optional, for established patterns)
└── changes/ # Proposed changes
├── [change-name]/ # Descriptive change identifier
│ ├── proposal.md # Why, what, and impact
│ ├── tasks.md # Implementation checklist
│ ├── design.md # Technical decisions (optional)
│ └── specs/ # Complete future state
│ └── [capability]/
│ └── spec.md # Clean markdown (no diff syntax)
└── archive/ # Completed changes
└── YYYY-MM-DD-[name]/
Change Storage Convention
Future State Storage
WHEN creating a change proposal THEN store the complete future state of affected specs AND use clean markdown without diff syntax
The changes/[name]/specs/ directory SHALL contain:
- Complete spec files as they will exist after the change
- Clean markdown without
+or-prefixes - All formatting and structure of the final intended state
Proposal Format
WHEN documenting what changes THEN the proposal SHALL explicitly describe each change:
**[Section or Behavior Name]**
- From: [current state/requirement]
- To: [future state/requirement]
- Reason: [why this change is needed]
- Impact: [breaking/non-breaking, who's affected]
This explicit format compensates for not having inline diffs and ensures reviewers understand exactly what will change.
Change Lifecycle
The change process SHALL follow these states:
- Propose: AI creates change with future state specs and explicit proposal
- Review: Humans review proposal and future state
- Approve: Change is approved for implementation
- Implement: Follow tasks.md checklist (can span multiple PRs)
- Deploy: Changes are deployed to production
- Update: Specs in
specs/are updated to match deployed reality - Archive: Change is moved to
archive/YYYY-MM-DD-[name]/
Viewing Changes
WHEN reviewing proposed changes THEN reviewers can compare using:
- GitHub PR diff view when changes are committed
- Command line:
diff -u specs/[capability]/spec.md changes/[name]/specs/[capability]/spec.md - Any visual diff tool comparing current vs future state
The system relies on tools to generate diffs rather than storing them.
Capability Naming
Capabilities SHALL use:
- Verb-noun patterns (e.g.,
user-auth,payment-capture) - Hyphenated lowercase names
- Singular focus (one responsibility per capability)
- No nesting (flat structure under
specs/)
When Changes Require Proposals
A proposal SHALL be created for:
- New features or capabilities
- Breaking changes to existing behavior
- Architecture or pattern changes
- Performance optimizations that change behavior
- Security updates affecting access patterns
A proposal is NOT required for:
- Bug fixes restoring intended behavior
- Typos or formatting fixes
- Non-breaking dependency updates
- Adding tests for existing behavior
- Documentation clarifications
Why This Approach
Clean future state storage provides:
- Readability: No diff syntax pollution
- AI-compatibility: Standard markdown that AI tools understand
- Simplicity: No special parsing or processing needed
- Tool-agnostic: Any diff tool can show changes
- Clear intent: Explicit proposals document reasoning