Workflow Recorder
Start setup →
🔥 Beta · open to the first 50 builders

Log your builds.

Workflow Recorder turns Claude Code and Codex Local sessions into activity replays: moving time, asks, commits, context load, and tool timing. Metadata only, never prompts, code, file contents, or screenshots unless you add an image yourself.

npx -y @workflowrecorder/buildlog@latest init --runtime codex
New here? Use the interactive setup guideClaude Code: npx -y @workflowrecorder/buildlog@latest init --runtime claudeCodex Local: npx -y @workflowrecorder/buildlog@latest init --runtime codex
Example replay detail
@buildwithmayaClaude Code + Codex Local

Morning build session

3h 42m moving · 12 asks · 4 commits shipped

12Prompts
148kTokens spent
42%Context load
54%Instant waits

Replay the work without exposing the work.

This sample uses fake data to show the public replay experience. Every public session gets a detail page with timing, tool counts, commits, context pressure, and approval/tool-wait timing, plus a profile surface where people can browse the sessions you choose to publish. The numbers stay measured, and the replay explains the activity minute by minute without exposing prompts, code, file contents, or screenshots.

How it works
1
Install once
Start with the interactive setup guide, pick Claude Code, Codex Local, or both, then run one command. The repo-local workflow-recorder plugin is optional for Codex users who want extra vocabulary and URL help. Both flows share the same account and config.
2
Keep coding
Every time Claude Code or Codex Local finishes responding, the matching Stop hook fires. Workflow Recorder parses metadata, computes stats, and uploads without prompts, dialogs, or "save?" interruptions.
3
Share when you want
Sessions are private by default. Visit your profile, tap "Make public" on the ones you want to share, copy the URL, drop it in Discord / Twitter / wherever. Lurkers see the replay detail without signing in.
What we see · what we never see
We capture
  • Session duration (active, excluding idle gaps > 30 min)
  • Tool counts: { Read: 12, Bash: 8, Edit: 5 }
  • Token counts per session (input / output / cache)
  • Number of files touched, number of commits made
  • Model used (Claude or GPT family)
  • Project + branch labels — shown only to you, never on public pages
We never capture
  • Your prompts or model responses
  • File contents or diffs
  • File paths
  • Bash command outputs
  • Tool call arguments
  • Screenshots or visual data unless you manually add an image URL
Ready? The fastest path is the guided setup: pick a runtime, run the installer, sign in, run doctor, then create one new response. start setup →