- Add bottom-up knowledge extraction pipeline (--mode knowledge)
- Configure thinking_budget=1024 for Gemini 2.5 models (4 min vs 24-80 min)
- Add JSON fallback parsing for LLM responses
- Add checkpointing and batch utilities for large repos
- Consolidate documentation into ARCHITECTURE.md, DEVELOPMENT.md, KNOWLEDGE_EXTRACTION.md

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

…covery

Major improvements to entity aggregation in knowledge extraction mode:

- Pass 1: Consolidate entities from component entities arrays (dedupe, merge fields, classify as core/supporting/integration/internal)
- Pass 2: Extract infrastructure from entity storage fields + external_dependencies (finds actual table/bucket names)
- Pass 3: Map service-to-service communication patterns

Key changes:
- Removed hardcoded patterns - now fully repo-agnostic
- Entity classification by LLM (core domain vs external integrations)
- Infrastructure extracted from entity storage fields containing actual resource names
- Updated WriteMarkdownDocs to render domain entities with tier badges and separate integration models section
- Updated documentation in KNOWLEDGE_EXTRACTION.md

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add _cleanup_extracted_tables() function for intelligent table filtering
- Add _parse_llm_json_array() helper for parsing LLM array responses
- Integrate LLM cleanup as Step 1.5 in AggregateEntities pipeline
- Update KNOWLEDGE_EXTRACTION.md with comprehensive documentation

The LLM cleanup pass is fully agnostic - no vendor names hardcoded:
- Filters out queues, buckets, loggers, search indexes
- Strips environment variable prefixes (process.env.xxx -> xxx)
- Deduplicates table variants
- Preserves parameterized names (${stage}-items)

Example: 62 raw tables -> 40 clean database tables

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Replace hardcoded "Lambda Functions" header with dynamic function_type
- Replace "AWS Resources" with dynamic infrastructure_type
- Replace "IAM Permissions" with generic "Permissions & Access Control"
- Update infrastructure extraction prompt to support multi-cloud (Lambda, Cloud Functions, Azure Functions)
- Make entity/table extraction examples generic (not DynamoDB-specific)
- Update infrastructure categories to use dynamic naming based on detected DB type
- Replace vendor-specific integration examples with generic categories

The generator now works equally well on any codebase type (backend, frontend, CLI, library) without assuming specific cloud providers or vendors.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Remove hardcoded Redux/Zustand/Context mentions from frontend extraction
prompts. The LLM now discovers whatever state management pattern the
codebase actually uses instead of assuming specific libraries.

Changes:
- repo_instructions: Generic "any state management pattern" instead of
  library-specific examples
- example_output: Generic "AppStore" instead of "Redux Store"

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add "Design Principle: Technology Agnostic" section to KNOWLEDGE_EXTRACTION.md
- Document frontend/CLI/library extraction categories
- Update README.md knowledge mode description to be generic
- Clarify that infrastructure is auto-detected, not vendor-specific

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Added escape_html_for_markdown() helper function that wraps HTML tags
in backticks to prevent Docsify from interpreting them as actual HTML.

Applied to all text content in _generate_component_chapter():
- Component descriptions and responsibilities
- File summaries
- Class/method/function purposes, params, returns, and logic
- Workflow details and data flow formats

Fixes rendering issues where tags like <style>, <defs>, <svg> would
break Docsify's markdown parser.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>