Produces a production-ready release specification document (Full or Lean format) including current-state audit, technical architecture with code examples, implementation plan, and risk assessment. Use when 'write a release spec', 'create a release specification for vX.X.X', or 'ground this spec in the codebase'.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Produces a production-ready release specification document (Full or Lean format) including current-state audit, technical architecture with code examples, implementation plan, and risk assessment. Use when 'write a release spec', 'create a release specification for vX.X.X', or 'ground this spec in the codebase'.
category
specification-driven-development
triggers
["write a release spec","create a release specification","ground this spec in the codebase"]
tier
1
agents
["primary"]
tool_dependencies
["file_system","bash"]
inputs
[{"name":"release_context","type":"string","description":"Description of what the release should accomplish (version, goals, scope)","required":true},{"name":"format","type":"string","description":"Specification format — Full, Lean, or Minimal","required":false}]
outputs
[{"name":"release_spec","type":"ref","format":"cas-ref","description":"Production-ready release specification document with current-state audit, architecture, implementation plan, and risk assessment"}]
Write Release Specification Skill
Version: 2.1 Created: 2026-02-02 Updated: 2026-02-07 Author: Manus AI Purpose: Write production-ready, A+ quality specifications for software releases
I. The Philosophy: Specification as Contract
A specification is not documentation—it is a contract. A vague specification invites confusion, rework, and failure. A rigorous specification is an act of respect for the builder's time.
Specifications created with this skill are:
Comprehensive: Every question the builder might have is answered
Precise: Technical details are specific, not hand-wavy
Actionable: The path from specification to implementation is clear
Testable: Success criteria are binary and measurable
The standard: 111/100 (A+). Good enough is not good enough.
II. When to Use This Skill
Use when:
Planning a new software version or release with multiple features
Designing a complex system architecture requiring detailed documentation
Commissioning work to an autonomous agent (Claude Code, etc.) that needs complete context
Coordinating parallel development tracks
Do not use for:
Small bug fixes or minor tweaks (use a simple task description)
Exploratory prototypes (use scouting first)
Features still being actively designed (finish scouting first)
III. The Workflow
Decision Point: Full Template or Lean Format?
Use the Full Template (Section IV) when: System is new, audience includes stakeholders, multiple teams implement, risk is high.
Use the Lean Format when: Architecture is established, audience is the implementing agent, scope is well-defined.
Lean Format structure: Route layouts, component tables, behavior lists. No preamble. "Sonnet level chunks" — direct, precise, implementable.
Rule: Match format to scope. Don't default to full template when lean will do.
Step 1: Gather Context and Inspiration
Read previous specifications — Study 2-3 recent specs for pattern and quality bar
Review the codebase — Use to understand current architecture
/repo-context-sync
Identify the problem — What pain point or strategic goal is this release addressing?
Scout alternatives — Use /strategic-scout if choosing between approaches
Step 1.5: Run Current State Audit
Specs describe the delta from measured reality, not from assumptions. Run before writing.
CREATE TABLE [table_name] (
id TEXT PRIMARY KEY,
field1 TEXT NOT NULL,
created_at TIMESTAMPDEFAULTCURRENT_TIMESTAMP
);
CREATE INDEX idx_[field] ON [table_name]([field]);
Integration Points:
Integrates with [existing component] via [method]
Extends [existing pattern] from v[X.X.X]
Performance Considerations:
[Caching strategy or database indexing]
[Expected latency or throughput]
3.3 [Feature/Component 2] - Detailed Design
[Repeat structure for each major component]
4. Implementation Plan
4.1 Phased Approach
Phase
Duration
Focus
Deliverables
1
Week 1-2
[Focus area]
[Specific deliverables]
2
Week 3-4
[Focus area]
[Specific deliverables]
4.2 Week-by-Week Breakdown
Week 1: [Focus]
Task 1: [Specific, actionable task]
Task 2: [Specific, actionable task]
Success Criteria: [What "done" looks like for this week]
[Continue for all weeks]
4.3 Dependencies & Prerequisites
Required Before Starting: ✅ [Prerequisite 1] Parallel Work: [What can be developed simultaneously] Blocking: [What must complete before other work starts]
4.4 Testing Strategy
Unit: [Component to test] — Target coverage: [X]% Integration: [Integration point to test] E2E: [User flow to test] Performance: [Metric] — Target: [Number] Manual QA: [Scenario] — [Edge cases]
Monitoring & Alerts: [Metric]: Alert if [condition]
7. Documentation & Communication
Update user guide with [new feature]
Update API documentation
Prepare changelog with breaking changes and migration guide (if needed)
8. Appendices
8.1 Future Considerations
v[X+1] Candidates: [Features deferred to next release]
8.2 Open Questions
[Question needing resolution before or during implementation]
8.3 References
[Link to related spec]
[Link to GitHub issue or discussion]
---
## V. Best Practices
| Practice | Why | How |
|----------|-----|-----|
| Start with vision, not features | Features without vision are a list of tasks | Write the vision statement first — if you can't state why in one sentence, you're not ready |
| Write production-ready code examples | Pseudocode leaves room for misinterpretation | Code that could be committed; include types and error handling |
| Use realistic timelines | Underestimating leads to rushed work and debt | Use past releases as benchmarks; 1,000-line feature = 1-2 weeks |
| Document integration points explicitly | Most bugs happen at system boundaries | For every component, document how it connects to existing systems |
| Include risk mitigation from the start | Identifying risks after implementation is too late | Ask "What could go wrong?" during architecture phase |
| Make success criteria binary and testable | Ambiguous criteria lead to scope creep | "User can create a project" ✅ vs. "UI is intuitive" ❌ |
| Reference existing patterns | Consistency reduces cognitive load | "Follow the structure of `ComponentX`" |
---
## VI. Quality Checklist
Before finalizing a specification, verify:
**Vision & Goals**
- [ ] Vision statement is a single, compelling sentence
- [ ] Goals are specific, measurable, and achievable
- [ ] Non-goals explicitly stated to prevent scope creep
**Technical Architecture**
- [ ] Every major component has detailed design with code examples
- [ ] All API endpoints fully specified (method, path, request, response)
- [ ] Integration points with existing systems documented
- [ ] Performance considerations addressed
**Implementation Plan**
- [ ] Timeline is realistic based on complexity
- [ ] Week-by-week breakdown has specific, actionable tasks
- [ ] Testing strategy is comprehensive (unit, integration, E2E, performance)
**Risk & Documentation**
- [ ] Major risks identified with mitigation strategies
- [ ] Rollback procedure defined
- [ ] User and developer documentation needs documented
**If you cannot answer "yes" to all 13 questions, the specification is not ready.**
---
## VII. Common Pitfalls to Avoid
❌ **Vague Goals:** "Improve user experience" → ✅ "Reduce context loading time by 50%"
❌ **Missing Code Examples:** High-level description only → ✅ Complete, runnable code
❌ **Unrealistic Timelines:** "2 days for full backend" → ✅ "2 weeks with phased approach"
❌ **No Risk Assessment:** Assumes everything will work → ✅ Identifies risks and mitigations
❌ **Incomplete Testing:** "We'll test it" → ✅ Specific test cases and coverage targets
❌ **No Integration Points:** Treats feature as isolated → ✅ Documents how it connects to existing system
---
## VIII. Related Skills
- **`strategic-to-tactical-workflow`** — Complete workflow from scouting to implementation (this skill is Phase 6)
- **`frontend-from-backend`** — For frontend specs needing deep backend grounding
- **`implementation-prompt`** — For converting this spec into implementation prompts
- **`parallel-tracks`** — For splitting large specs into parallel execution tracks
- **`repo-context-sync`** — For gathering codebase context before writing specs
- **`memory-garden`** — For documenting learnings from implementation
---
**Last Updated:** 2026-02-07
**Maintained By:** Manus AI
**Status:** Active
---
## Output
- A release specification document saved to `docs/vX.X.X/[release]_specification.md`
- Includes a "Current State (Audit Results)" section with measured counts, not estimates
- Technical architecture section with production-ready Go and TypeScript code examples, API endpoint tables, and database schema
- A complete quality checklist appended confirming all 13 readiness criteria are met before commissioning
## Examples
**Scenario 1:** "Write a spec for DojoGenesis v0.2.5 adding the channel bridge and WebSocket hub." → A Full-format spec with a current-state audit measuring existing test files and API endpoints, detailed design sections for each new component with Go struct definitions and TypeScript interfaces, a two-week phased implementation plan, and a risk table covering WebSocket connection stability.
**Scenario 2:** "Spec the new entity CRUD feature for the v0.3.0 release." → A Lean-format spec with a route layout table (GET/POST/PUT/DELETE for `/api/v1/entities`), a component table mapping each UI widget to its handler, behavior list, test cases, and rollout notes — no preamble, ready to commission.
## Edge Cases
- When scouting is not complete and the architecture is still undecided, stop and complete the scout before writing the spec — a spec written from open architecture questions will need to be substantially rewritten
- When the spec covers work for parallel tracks, write a master release spec first and then use `spec-constellation-to-prompt-suite` to derive the per-track prompts
- When the codebase has changed since the previous spec, run Step 1.5 (current state audit) against the latest main branch — stale metrics produce grounding errors
## Anti-Patterns
- Writing the spec from memory of the codebase instead of running the actual audit commands — even a 5-day-old mental model will have drifted; always run the grep commands
- Skipping "Non-Goals" — implementation agents will fill undefined scope with adjacent features; explicit non-goals prevent this
- Marking a spec "Final" before running the 13-item quality checklist — the checklist exists precisely because authors miss items when reviewing their own work
