claude-replay
Converts AI coding agent sessions (Claude Code, Cursor, Codex, Gemini, OpenCode) into self-contained, embeddable HTML replays. Think screen recordings but for agent sessions -- interactive, searchable, and vastly smaller than video files. A single HTML file with no external dependencies.
Key Quotes
"Replay files embed the full session transcript, including source code, file paths, tool inputs/outputs."
Key Themes
This fills a genuine gap in the agent tooling ecosystem. Agent sessions produce JSONL transcripts that are functionally unreadable. claude-replay turns them into something you can share in a blog post, attach to a bug report, or use as teaching material. The fact that it supports five different agent formats (Claude Code, Cursor, Codex, Gemini, OpenCode) makes it a universal tool rather than a Claude-specific one.
The architecture is pragmatic: parse JSONL, compress with deflate, embed in vanilla JavaScript that decompresses via the browser's native DecompressionStream API. No React, no build system, no runtime dependencies. Size reduction of 60-70% means these are genuinely shareable.
The live watch mode is interesting for a different use case: monitoring running agent sessions in real-time. Combined with the web editor for visual session editing, this starts to look like a proper observability tool rather than just a replay generator.
Critical Analysis
The automatic secret redaction (API keys, AWS credentials, bearer tokens) is essential but explicitly "best-effort." This is honest but means you should never blindly share a replay without reviewing it first. The tool itself warns about this.
The bookmark/chapters system addresses a real problem -- agent sessions can be very long, and navigating to the interesting part is important. But the bookmarks are manually added via --mark "N:Label". What you really want is automatic detection of significant moments (test failures, tool errors, mode switches).
See also session-analysis for Leonard Lin's approach to understanding session dynamics, and agent-pr-replay for comparing agent vs human approaches to the same PR.
Sources: claude-replay Last updated: 2026-05-14