llm‑echo 0.5a0 Adds -o thinking 1 Flag for Fake Reasoning Blocks
Explore the `-o thinking 1` flag in llm‑echo 0.5a0 to simulate reasoning blocks during automated tests.
Run the example command with `-o thinking 1` to verify that the echo model outputs a reasoning block and JSON echo as expected.
Summary
The llm‑echo plugin has reached version 0.5a0, adding a new `-o thinking 1` flag that simulates a reasoning block when testing against LLM 0.32a0 and newer. The echo model is a lightweight stub that returns the prompt as JSON without invoking an actual LLM, making it ideal for automated test suites. With the new flag, the plugin writes a fake reasoning block to standard error before outputting the JSON echo, giving developers a clearer view of the internal flow. The release also includes documentation updates that explain how to enable the flag and interpret its output.
To use the feature, developers run `uvx --with llm==0.32a1 --with llm-echo==0.5a0 llm -m echo hi -o thinking 1`. The command triggers the echo model, prints a reasoning block to stderr, and returns a JSON object that echoes the prompt. This enhancement streamlines testing by allowing teams to verify that their code handles reasoning blocks correctly without needing a real LLM backend.
Key changes
- Introduced `-o thinking 1` flag for fake reasoning block
- Echo model returns prompt as JSON without LLM execution
- Reasoning block printed to standard error
- Example usage shown in docs
- Updated docs to explain flag and output interpretation