llm-echo 0.5a0 adds thinking-block simulation for LLM 0.32a0 testing

Simon Willison shipped version 0.5a0 of llm-echo, a plugin that registers a fake “echo” model inside the LLM CLI tool. The model performs no inference — it simply returns the prompt back as JSON — which makes it a stable fixture for writing automated tests against LLM-based code without burning tokens or depending on a real provider.

The new release adds a -o thinking 1 option aligned with LLM 0.32a0’s reasoning support. When set, echo emits a synthetic reasoning block to stderr before the JSON response, letting test suites exercise the thinking-block code paths introduced by the recent 0.32a0 refactor. The plugin is invokable on demand via uvx --with llm==0.32a1 --with llm-echo==0.5a0.