Release v2.15.0 - Full/Sliced File Reading & Large File Summarization · spelech/contextcortex

What's New in v2.15.0

Safe Path Resolution: Built FileReaderService supporting secure path resolution across both monitored local directories (indexed_paths) and uploaded local storage files (/data/storage), with robust path traversal protection.
Line-Range Slicing: Slices file contents with 1-based, inclusive line bounds (start_line, end_line), capping lines at read_file_max_lines (default: 2,000 lines) with clear truncation indicators.
Binary Detection: Automatically checks for null bytes in initial bytes to reject binary files gracefully.
MCP Tool & API: Exposed read_file via FastMCP tool and REST endpoint (GET /admin/api/files/read).

Ingest Auto-Summarization: Automatically generates structured LLM markdown summaries during indexing for files exceeding configurable threshold (default: 500 KB) up to max size (default: 10 MB).
On-Demand Summarization: Retrieve cached summaries or regenerate summaries via summarize_file tool and POST /admin/api/files/summarize.
Vector Search & SQLite Caching: Persists summaries in SQLite file_summaries.summary_text and indexes them into vector store as doc_type="summary" documents.

File Management & Summarization Settings: Added a new dedicated configuration panel in Settings view to configure summarization thresholds, max file size, line limits, and dedicated LLM summary model.
API Endpoints: GET /admin/api/settings/files and POST /admin/api/settings/files.

Eliminated uncontrolled path access patterns in processor early-return checks.
Sanitized exception error handling across file and settings endpoints to eliminate information exposure risks.