llmflow 3.0.2

CRAN resubmission addressing reviewer feedback:

• Replaced cat() with message() for suppressible console output
• Replaced globalenv() with isolated evaluation environment to comply with CRAN policy
• Code execution now uses .llmflow_eval_env within callr sessions, never modifying user’s global environment

llmflow 3.0.1

Initial CRAN submission.

Features

• AutoFlow(): Complete workflow combining retrieval-augmented generation (RAG) with ReAct execution
• react_r(): ReAct loop implementation with smart history management and context summarization
• response_to_r(): LLM-to-R code generation and execution in isolated sessions
• response_as_json(): Structured JSON output with schema validation
• retrieve_docs(): R function documentation retrieval for RAG workflows

Architecture

• Dual-LLM support for optimized cost and performance
• Error escalation and task degradation strategies
• Intermediate result protection near turn limits
• Session continuity via existing_session parameter

Notes

• Functions extract_package_docs(), extract_function_docs(), extract_docs_from_tarball(), and generate_qa_from_docs() have been moved to the separate llmtools package.