Jonah Schulte
|
73b8190e7b
|
feat(validation): Add comprehensive story validation system with Haiku agents
VALIDATION WORKFLOWS (6 total):
- validate-story: Quick task checkbox validation
- validate-story-deep: Deep code verification with Haiku agent
- validate-all-stories: Batch quick validation
- validate-all-stories-deep: Comprehensive platform audit
- validate-epic-status: Per-epic validation
- validate-all-epics: All epics validation
VALIDATION SCRIPTS (4 total):
- sprint-status-updater.py: Compare story files vs sprint-status.yaml
- task-verification-engine.py: Python-based task verification
- llm-task-verifier.py: LLM-powered verification (alternative)
- add-status-fields.py: Add Status field to stories
HAIKU AGENT APPROACH:
- One agent per story (not per task - avoids 99% overhead)
- Agent reads actual code with Glob/Read tools
- Verifies stubs vs real implementation
- Checks multi-tenant, error handling, tests
- Evidence-based verification (line numbers, code snippets)
COST OPTIMIZATION:
- Haiku: $0.15/story vs Sonnet: $1.80/story (92% savings)
- Full platform: $76 vs $920 (saves $844)
- Batching: 5 concurrent agents (prevents overload)
CAPABILITIES:
- False positive detection (checked but code missing)
- False negative detection (unchecked but code exists)
- Code quality review (TODOs, stubs, missing features)
- Status recommendation (done/review/in-progress)
- Automated status updates
DOCUMENTATION:
- HOW-TO-VALIDATE-SPRINT-STATUS.md
- SPRINT-STATUS-VALIDATION-COMPLETE.md
- Slash command docs in .claude-commands/
USE CASES:
- Weekly: Quick validation (free, 5 sec)
- Pre-done: Deep story check ($0.15, 2-5 min)
- Pre-launch: Full audit ($76, 4-6h)
- Quality sweep: Phase 3 comprehensive validation
Enables bulletproof production confidence for any BMAD project.
|
2026-01-02 19:20:11 -05:00 |