Agent PR Replay takes merged PRs from any repository, reverse-engineers the task prompt, runs Claude Code against it, and compares what the agent did versus what humans actually shipped. The result is targeted, empirical guidance.

No description available

Session replay → UX anomaly detection → RAG → multi-agent fixes → draft GitHub PR (PostHog-shaped demo)

take an examplar git repo with known PR's and replay them letting PR agents run against them. Then gather up the issues and label for evaluation of PR tools

ReplayCI: PR-native regression tests for tool-using AI agents with deterministic replay, behavior diffs, and cost/safety gates.

No description available