- Generate prompts/phase{1-4}-prompt.txt files with phase-specific tasks
- Update GUIDE.md with phase-by-phase execution instructions
- Include both /ralph-wiggum and bash loop alternatives for each phase
- Each phase prompt references PROMPT.md and focuses on one phase only
- Add tech-stack-specific build/test/lint commands per phase
Also includes fix for JSON parsing with markdown code fences.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add Claude API client with Anthropic SDK
- Retry logic with exponential backoff (3 attempts)
- Handle auth errors, rate limits, and server errors
- Architecture classification with JSON parsing
- Research query generation from architecture
- Add Perplexity API client with native fetch
- Retry logic with exponential backoff
- Parse responses with sources and citations
- Execute multiple research queries
- Add prompt templates for architecture classification
- Add default query generation as fallback
- Add ArchitectureGenerator combining Claude + Perplexity
- Add 44 new tests (80 total, all passing)
All Phase 2 acceptance criteria met:
- Claude client sends prompts and retries on failure
- Perplexity client searches and parses with sources
- Architecture classification returns valid JSON
- Research queries generated (4-6 per architecture)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
