pub fn build_reflection_prompt(
user_request: &str,
agent_response: &str,
quality_signals: &QualitySignals,
tool_errors: &[String],
) -> StringExpand description
Build the reflection prompt that asks the LLM to analyze a suboptimal turn.
This is the pure prompt-construction step for reflection generation:
- the original user request becomes the task context,
- the agent response becomes the failed/suboptimal attempt,
- tool errors and quality signals become the environment feedback,
- and the output contract forces the model into the structured
ATTEMPT/ISSUE/STRATEGY/TAGSformat expected by the parser.