Changelog

All notable changes to the rlm-rs Claude Code plugin will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

batch_size argument to /rlm-query command for configurable subcall batching (default 10)
argument-hint frontmatter to all commands for improved autocomplete display
Document index field in search results for temporal/document ordering of chunks

Increased default --top-k from 10 to 100 for chunk retrieval to improve analysis coverage
Updated default chunk size to 6000 characters (~1500 tokens) for better semantic search balance
Reduced default batch size from 20 to 10 chunks per subcall agent

Technical overview documentation explaining RLM implementation, hybrid search, and architecture

Core RLM Workflow: Full implementation of the Recursive Language Model pattern from arXiv:2512.24601
Hybrid Search: BM25 + semantic search with Reciprocal Rank Fusion (RRF) for finding relevant chunks
Pass-by-Reference: Chunk retrieval by ID instead of file-based workflow for efficiency
Commands:
- /rlm-init - Initialize SQLite database
- /rlm-load - Load files with chunking (fixed, semantic, parallel)
- /rlm-status - Show current buffers and state
- /rlm-query - Run queries with automatic chunk analysis
Skills:
- rlm - Main workflow orchestration for large context processing
- rlm-chunking - Chunking strategy selection guide
Agents:
- rlm-subcall - Haiku-based chunk analyzer with JSON output
- rlm-synthesizer - Sonnet-based result aggregator
Documentation:
- Architecture documentation with diagrams
- CLI reference guide
- Contributing guidelines