rishabhpande /
Project URL Category Python Project Title Replay bad retries before they hit real tools Project Description A local Python lab for failed agent retries. Replay duplicate writes, partial updates and stale versions, then i…
Project URL Category Python Project Title Replay bad retries before they hit real tools Project Description A local Python lab for failed agent retries. Replay duplicate writes, partial updates and stale versions, then inspect the saved state. Beginners can compare two scripted strategies without a model key. Highlights I maintain this MIT-licensed project. The teaching hook is a lost acknowledgement: a comment was saved, but its caller saw a timeout. Retrying blindly creates a duplicate. The replay lets readers step through the calls and compare what the agent claimed with what was actually persisted. The quick start generates a self-contained HTML replay with eight scripted runs across four fault scenarios. It runs offline after installation, with synthetic data and no API key. Readers can add YAML task fixtures or write a local Python adapter, then test it against final-state assertions. The two included strategies are deterministic teaching baselines, not LLMs, and the demo is not a measured LLM benchmark. Custom adapters execute trusted local Python; this is not a sandbox or a production exactly-once guarantee. Example Code sh python -m pip install -e . retry-rodeo demo Run inside the cloned repository, then open rodeo-output/index.html. Screenshots or Demo Videos Public interactive replay: I checked the HelloGitHub website search for the repository URL and the repository's issue history before submitting; no prior recommendation was found.
Discuss this product