Ray5th/Lean4-helper
Minimalist UI pass: remove emojis and decorative chrome
Repo cleanup: dead code out, test isolation fix, real README
RAFT-style prompting: distractor-aware framing + usage citations
Retrieval fixes: robust nprobe tuning + goals-only query
Polish UI: labeled controls, depth, live solving feedback
Disable two-stage retry + warn on mid-generation model switch
Fix Groq model IDs against live catalog
Refresh Groq model lineup — drop Llama 3.x/Gemma, add GPT-OSS + friends
Fix UI bugs: bounded panel heights + readable text everywhere
Swap MiniLM retriever for LeanDojo's ByT5 premise encoder
Pin ruff to 0.15.14 and re-sort imports
Add CI/CD: GitHub Actions workflow + ruff config
Add Claude CLI provider for Pro-token billing
benchmark.py: support Claude API + local-files mode
Speed up proof generation: cache components, smaller max_tokens, two-stage retry
Add Claude API support (users supply their own key)
Redesign UI: light Claude.ai-inspired theme
Merge pull request #8 from ray5th/worktree-agent-a0c1e62025fb04462
Merge pull request #7 from ray5th/worktree-agent-a5d6c683e40e73f66
Merge pull request #12 from ray5th/claude/test-scripts-clis
Merge pull request #11 from ray5th/worktree-agent-ab4d84b9b8326f21b
Merge pull request #6 from ray5th/worktree-agent-a4b681bccbbb9eb5f
Merge pull request #15 from ray5th/fuzz/solve-proof-corpus
Merge pull request #5 from ray5th/claude/festive-bouman-4a86a7
Fix 4 bugs found in manual code review
Fix 4 bugs found by stress-test agents + add fuzz/concurrent test suites
Add unit tests for run_agent and build_index CLIs
Add unit tests for ProofAgent wrapper
Add pre-built Mathlib FAISS index (118,885 declarations)
Gracefully skip RAG when FAISS index is missing
Fix stdout race in solve_proof under concurrent requests
Fuzz retriever query handling (fix missing input guards)
Fuzz state machine + fix empty-LLM-output bug
Add parse tests for every problems/*.lean file
Redesign UI as two-pane VSCode-style editor
Redesign Gradio UI with VSCode-style Claude theme
Remove Lean+Mathlib warmup step from Dockerfile
Add missing langchain-huggingface and langchain-classic packages
Fix Dockerfile: collapse multi-line python -c into single line
Fix HF Spaces colorFrom value (orange not in allowed list)
Add Hugging Face Spaces YAML frontmatter to README
Fix Vercel CI: serve static redirect page instead of Python function
Add Hugging Face Spaces deployment with Groq API
Merge pull request #4 from ray5th/claude/intelligent-varahamihira-81b22c
Resolve merge conflicts with main, keeping improved versions
Add MiniF2F benchmark harness and improve proof agent robustness
Fix LLM hallucinated imports and slow thinking mode (#3)
Fix LLM hallucinated imports and slow thinking mode (#2)
Fix LLM hallucinated imports and slow thinking mode
Fix import paths and Mathlib auto-detection for LangChain 1.x
