[spelech]

Summary

• Default ripgrep (rg) engine across grep and indexing (ripgrep_runner.py, code_grep_scanner.py, grep_handlers.py, local_syncer.py, git_delta_scanner.py): Replaces Python os.walk + line-by-line re.search scans with streaming rg --json and rg --files, including PCRE2 fallback for lookarounds, broken symlink safety, and pure-Python fallback when rg is unavailable. Removes the duplicated 100-line extended os.walk loop from grep_handlers.py.
• Word-token query classification & t=0 parallel Track A / Track B search (query_router.py, lexical_retriever.py, search.py):
  • Distinguishes exact identifier/path/quoted queries (lexical_exact), agent keyword-series queries such as "order processor retry payment" (lexical_keywords), and conceptual natural-language questions (semantic_prose) using word-token morphology and prose stopword/interrogative density.
  • Runs Track A (tokenized AST symbol lookup + rg multi-token cluster scan + file summaries) concurrently at t=0 alongside Track B (embedding + vector search).
  • Applies a word-token coverage confidence gate to return Track A immediately without waiting for slow embeddings on exact and high-confidence keyword-series queries, while awaiting and fusing Track A + Track B on conceptual/ambiguous queries.
• Whole-function AST chunk preservation (state.py, processor.py, symbol_extractor.py): Introduces CODE_CHUNK_SIZE=4000 (up from 1500) so functions and methods stay intact in single AST chunks.
• Windows portable rg.exe bundling (build_portable_windows.ps1, build-portable-windows.yml): Downloads and packages rg.exe alongside contextcortex.exe with CI smoke verification.

Verification

• pytest -q: 678 passed, 16 skipped (including new tests in tests/test_ripgrep_and_query_router.py and style guardrails <= 500 LOC / semantic naming checks).