trajectory-replay-demo
What it is
A Python tool for recording agent tool-call trajectories to SQLite and replaying them against different models or prompts to measure planning quality. It uses stubbed tool results to allow cheap A/B testing without executing live tools.
The problem it solves
Testing agent prompt or model changes traditionally requires re-running expensive, slow, and non-deterministic end-to-end traces with live tools, making it difficult to isolate planning quality improvements.
What you could use it for
- A/B testing agent system prompts to see if they improve tool selection
- Evaluating the impact of swapping LLM models on agent planning consistency
- Reproducing agent behavior deterministically for debugging
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.