Crate llm_repair 

llm-repair

Utilities for repairing malformed JSON output from LLMs and extracting tool calls from unstructured text.

It is designed to be robust against common failure modes of smaller models, such as truncation, invalid escapes (e.g. LaTeX), and “conversational” formatting.

Re-exports

pub use extraction::extract_evaluations_from_markdown;

pub use extraction::extract_proposal_from_markdown;

pub use extraction::extract_python_tool_calls;

pub use extraction::extract_xml_tool_calls;

pub use extraction::heuristic_json_tool_calls;

pub use history::pair_orphan_tool_calls;

pub use history::stub_tool_response;

pub use repair::repair_aggressive_escapes;

pub use repair::repair_conversational_response;

pub use repair::repair_invalid_escapes;

pub use repair::repair_tool_calls;

pub use repair::repair_truncated_json;

pub use repair::sanitize_json_string_lossy;

pub use utils::clean_json_string;

Modules

extraction

Functions for extracting structured tool calls from unstructured text.

history

Conversation-history hygiene.

repair

Functions for repairing broken or malformed JSON strings.

utils

Utility functions for string manipulation and cleanup.