Runner, a local LLM engine, handles tool calls that are truncated by token limits by completing them as valid JSON and returning a parseable result, allowing agent loops to continue instead of failing or retrying from scratch like other engines.