quell reproduce
quell reproduce turns a bug description in plain English into a pytest test that fails while the bug exists. Use it when someone reports a bug and you want a regression test before you fix it. According to the project README it needs an LLM provider configured with quell auth.
What is the syntax?
quell reproduce "payment accepts zero amount"
quell reproduce "payment accepts zero amount" --file src/payments.py
| Argument or option | Meaning |
|---|---|
DESCRIPTION | The bug in plain English. Required. |
--file PATH | Target source file. |
--root PATH | Project root (default .). |
What happens?
The bug reader parses the description (with the LLM), the synthesis step generates a test for the bug, and the verification pipeline checks it. For bug reproductions no violation is injected: the source is already considered broken. On success the command prints a panel:
Bug reproduction test written.
The test currently FAILS (bug exists). Fix the code, then run it to confirm.
If no verified test could be generated it prints Could not generate a verified bug reproduction test. and exits with code 1.
What does it send to the LLM?
The bug description is parsed with an LLM prompt, and test generation uses the prompt described in LLM providers. Rule-based commands such as quell find never call an LLM unless you pass --use-llm.
Can an agent do this?
Yes: the MCP server exposes reproduce_bug(description, file?). See MCP server and plugin.