| name | tasks-feature-lifecycle |
| description | Feature lifecycle: requirements, discovery, design approval, quality gates, production.
Use when: planning feature lifecycle from requirements through design approval to production;
crystallizing vague requirements into actionable specs.
NOT for: bug fixes, refactoring, or code review without new feature context.
|
Feature Development
Guided feature implementation with discovery, design approval, and quality gates.
Quick Reference
| Intent | Action |
|---|
| Understand request | Parse requirements, clarify, confirm |
| "Vague request", "make it better" | CLARITY framework: clarify all ambiguities |
| "Break this down" | Task breakdown with milestones |
| Map codebase | Grep for patterns, Glob for files, Read key files |
| Design options | Present 2-3 architectures with trade-offs |
| Implement | Wait for user approval, then edit files |
| Validate | Run tests, lint, build |
MUST (BLOCKING):
- Read files before Edit (violate = hallucinated code)
- Run quality checks before Summary (violate = incomplete)
NEVER:
- Skip Clarification phase (ambiguity = rework)
GATE behavior:
- When running interactively (root agent): pause and ask user at each GATE
- When running as subagent (inside Task): self-assess gates using available context — document assumptions made, flag low-confidence decisions in output for user review
Workflow
- Discovery: Parse feature request, summarize understanding
- GATE: user confirms understanding (subagent: document understanding, flag assumptions)
- Exploration: Grep/Glob/Read to map similar features, architecture, integration points
- 3 parallel searches: similar features, architecture, integration
- Clarification: Identify ALL ambiguities (edge cases, error handling, scope, compatibility)
- GATE: all questions answered (subagent: list unresolved questions in output, proceed with best-guess defaults)
- Design: Present 2-3 architecture options with trade-offs
- GATE: user selects approach (subagent: recommend top option with rationale, note alternatives)
- Implementation: Edit files, verify build
- Interactive: wait for explicit approval first
- Subagent: proceed with recommended design, document choice
- Review: Check for issues, report at >=80% confidence
- GATE: user decides which fixes to apply (subagent: apply fixes above 90% confidence, list others)
- Summary: Report files created/modified, key decisions, assumptions made, next steps
Quality Checklist
[ ] All acceptance criteria met
[ ] Tests pass (unit + integration)
[ ] Lint/typecheck pass
[ ] Build succeeds
[ ] No regressions introduced
[ ] Code follows existing patterns
Memory Safety
| Limit | Value | Rationale |
|---|
| Files read in exploration | 30 | Focus on relevant patterns |
| Architecture options | 3 | User can't evaluate more |
| Review issues shown | 10 | Prioritize by confidence |
Anti-Patterns
| Anti-Pattern | Problem | Fix |
|---|
| Skip clarification | Deliver wrong feature — rework required | MUST answer ALL ambiguities before design |
| Implement before approval | User rejects approach — wasted work | Wait for explicit "proceed" after design |
| Too many options | User paralysis — decision fatigue | Present top 2-3 with clear trade-offs |
| Generic exploration | Slow, unfocused — irrelevant results | Use targeted Grep/Glob for discovery |
| Skip quality gates | Broken build — incomplete delivery | Run verification before summary |
Examples
Before: Skipping Clarification
User: "Add user authentication"
Claude: [immediately starts coding JWT middleware]
Result: Wrong approach - user wanted OAuth, not JWT
After: Enforcing Clarification
User: "Add user authentication"
Claude: "Clarifying requirements before design:
1. Authentication method? (OAuth, JWT, session-based)
2. User storage? (database, external provider)
3. Session duration? (remember me, logout behavior)
4. Password reset flow needed?"
[waits for answers before design phase]
Edge Cases
| Situation | Behavior |
|---|
| User unclear on requirements | Ask clarifying questions, don't assume |
| Feature overlaps existing code | Map overlap, propose integration |
| User says "just do it" | Provide recommendation, confirm once |
| Scope creep during implementation | Note additions, ask if should include |
| Existing patterns conflict | Document conflict, ask user preference |
Success Criteria
Related Skills
- Skill: tasks-spec-methodology - Transform rough notes into developer-ready specs (pre-Phase 1)
- Skill: next-audit-methodology - Deep dive into existing patterns (use before Phase 2)