RLM (Recursive Language Model) Skill

To analyze large log files exceeding the LLM context window, you load the text as an external variable, chunk it into manageable pieces, run per-chunk sub-queries, and aggregate the partial results into a synthesized response.

Recursive chunking is a technique that breaks down massive content into smaller segments by lines, characters, or paragraphs, allowing an LLM to process each chunk individually and extract cross-chunk patterns without losing information.

Yes, you can summarize a massive codebase without hitting token limits by loading the content outside the prompt, applying a chunking strategy, and using batched sub-queries to extract information like TODOs before aggregating the outputs.

You extract cross-chunk patterns across multiple documents by running batched sub-queries concurrently on each chunk, then using an aggregation step to synthesize the partial results into a single unified response.

The best way to inspect a large file structure before processing is to load it as an external variable and retrieve metadata and previews, allowing you to understand the content before running recursive sub-queries.

While chunk-based analysis handles 10M+ token workloads by processing segments individually, the final response must be synthesized from many partial results, meaning the aggregation step must effectively summarize extensive sub-query outputs.