Orchestrate long-context processing with a root LLM, persistent REPL, and subagent for chunk analysis.
npx skills add https://github.com/Ju-usc/claude_code_rlm --skill rlm-ju-usc
Please help me install this Agent Skill. Skill: rlm Source: https://github.com/Ju-usc/claude_code_rlm/tree/main/.claude/skills/rlm Command: npx skills add https://github.com/Ju-usc/claude_code_rlm --skill rlm-ju-usc
Processing long-context content by orchestrating recursive LLM interactions with a persistent local REPL and a subagent to handle chunk analysis.
Initialize the REPL with your large context file, then start the recursive analysis workflow to chunk, analyze, and synthesize results.
High-intent search queries and answers about installing and using this skill.
To process large documents and logs, you can use a chunking workflow that splits the content, delegates chunk-level analysis to a subagent, and synthesizes the aggregated results through a persistent REPL.
A persistent REPL maintains state and buffers across multiple LLM invocations, allowing the root orchestrator to coordinate recursive chunk processing and aggregate structured outputs without losing context between steps.
The best way to analyze long transcripts is to generate chunks, delegate each chunk to a subagent for localized analysis, and aggregate the synthesized sub-results into a final structured output via stateful orchestration.
Yes, you can use a Python REPL to maintain persistent state and buffers across recursive LLM interactions, enabling continuous chunk generation, subagent delegation, and structured aggregation for large files.
Long-context chunking works for massive text files by dividing the input into manageable chunks, using a subagent to search and analyze each segment, and then synthesizing the findings into a unified output.
You should use a subagent for chunk-level analysis when documents are too large for a single context window, requiring stateful chunk generation, isolated chunk processing, and final structured aggregation to avoid context loss.
blueskies1818Execute Python code with persistent state via a live Jupyter kernel.
charleshall888Dispatch parallel reviewer agents to synthesize objections into a structured critique.
ivalmartChunk long-context documents and delegate analysis to an rlm-subcall subagent.
brainqub3Execute recursive language model loops for long-context tasks via a local Python REPL.
