Crackhaus
Marionette, our project addresses both statement 1 and 3.
Core trick
A CLI wrapper scores each user prompt for vagueness and detects agent sycophancy, then halts and reruns the response with a corrective instruction.
Description
Marionette, our project addresses both statement 1 and 3.
Marionette is a cli tool that lives on top of claude code and other ai agents, which monitors both the user and the agent. The user is monitorred for poor prompts, and marionette encourages the user to write more detailed ones. The agent is monitored for sycophancy, and marionette automatically stops the agent, then forces the agent to rerun, but with less sycophancy.
The linked repo returns 404, so it was deleted or made private.