text-only-nudge

To stop plan-only replies and force tool execution, you can inject a text nudge into the agent loop. This mechanism detects descriptive behavior during early iterations and pushes a prompt to trigger immediate action.

LLM text-only responses during the first few iterations often occur when the model outlines a plan but lacks a trigger to act. A nudge injection identifies this state and prompts the agent to run commands immediately.

Yes, you can force tool execution during early iterations by applying iteration control logic. This nudge specifically targets initial iterations, checking for action words and the absence of prior tool calls before injecting its prompt.

Text nudge injection is a safety and iteration control mechanism that pushes models to execute tools. It detects when a model generates content without prior tool calls and injects a system prompt to enforce immediate execution.

Plan-only replies are detected by checking for model content combined with a lack of prior tool calls. If specific action words are present, a nudge message is injected to force the agent to start executing tools.