ypollak2/llm-router

Major release: Automatic quality scoring, developer-first README rewrite, documentation consistency pass, stale asset cleanup.

Added

• Quality Feedback Loop (Sprint 4) — src/llm_router/quality_feedback.py
  • Auto-scores every routed response using content heuristics (code blocks, structure, refusals, citations)
  • Per-model quality tracking with minimum-calls threshold (3) before trusting signal
  • should_skip_model() — routing engine skips models with avg quality < 0.4 for specific task patterns
  • Integrated into router.py dispatch loop and all text.py tools (query, research, generate, analyze, code)
  • 23 new tests in tests/test_quality_feedback.py

Changed

• Complete README rewrite — developer-first, text-based, high-trust landing page
  • No images/SVGs — shields.io badges only
  • Honest "Use this if / Don't use this if" section
  • Accurate tool count (60 MCP tools), package names, provider list
  • ASCII architecture diagram, markdown tables throughout
• Documentation consistency pass — corrected "48 tools" → "60 tools" across 10+ docs
• Package name corrections — pip install llm-routing consistently referenced
• Tool count standardized — 60 MCP tools (56 llm_* + 4 agoragentic_*) across all docs

Removed

• 18 orphaned SVG assets from docs/readme/ (stale claims, zero references)

Fixed

• SECURITY.md referenced wrong package name (claude-code-llm-router → llm-routing)
• HOST_SUPPORT_MATRIX.md referenced wrong install command
• server.py and docs/TOOLS.md had outdated tool counts