v0.7.3 — is_implementation Heuristic in Planner & Token-Aware File Skip

What's new

is_implementation heuristic in planning summary

• _build_planning_summary() now counts impl_file_count, test_file_count, config_file_count using the same path heuristic as the RAG embedder
• LLM planner receives these counts to make more precise wiki structure decisions:
  • High impl_file_count → more core-component pages
  • test_file_count < 3 → skip dedicated testing page
  • config_file_count < 2 → skip configuration page

Token-aware file skip (env var configurable)

• CLOSE_WIKI_MAX_CODE_CHARS (default: 320000 = ~80K tokens)
• CLOSE_WIKI_MAX_DOC_CHARS (default: 32000 = ~8K tokens)
• Embedder pre-checks file size, logs and skips oversized files
• Progress callback reports: "Embedding 42 chunks from 8 files (2 files skipped — too large)…"

Tests

116 passed, 3 warnings
Coverage: 74.92%