Claude Code can run sessions two ways: locally (the CLI on your machine) or in the cloud — sessions started from claude.ai/code (the web UI, or the "Cloud" / "Environment" launch options in the desktop app). Cloud sessions run in an Anthropic-hosted sandbox and leave no trace on your machine — nothing is written to the local logs that /pulse-capture normally scans.

As of /pulse-capture v1.1, this is handled automatically. You don't open a browser console, run any JavaScript, or paste anything — the skill fetches your cloud session list itself and folds it into the same proposal as your local work.

What you need

Nothing to configure. Just make sure a browser context is connected and signed in to claude.ai when you run /pulse-capture:

• Claude in Chrome connected, or
• An in-app browser pane, if your Claude Code setup has one — signed in to claude.ai.

No browser available? The skill still runs — it explicitly tells you cloud work is missing from the proposal, rather than silently producing a local-only timesheet.

How it works (for the curious)

1. The skill opens https://claude.ai/code and calls the same endpoint your browser already uses to list your sessions, authenticated with your existing claude.ai login — not a separate credential.
2. It keeps only what it needs — id, title, timestamps, source repos, branch names — never the raw response. The raw response includes a resume_token, which is a credential, not session metadata; the skill never touches it.
3. The result feeds scripts/cloud_blocks.py, which:
   • Caps any single session at 240 minutes by default, so a tab left open for days can't claim days of "work." Anything hitting the cap is flagged for your review, not logged automatically.
   • Merges overlapping sessions into real wall-clock spans — if you kick off one cloud session and start another while it's still running, your total isn't double-counted.
4. Cloud blocks merge with your local blocks into a single proposal. Nothing reaches Pulse until you approve — same as every other /pulse-capture entry.

Sanity-checking cloud time

A cloud session's last_event_at isn't proof you were actively working — an idle tab can drift forward with no real activity behind it. The skill can't independently verify every session against your repo history, so:

• Blocks the cap catches are flagged, not silently trusted.
• If a proposed entry's duration looks off, edit it before approving. What you approve is what lands on your timesheet — not whatever the API happened to report.

Privacy

Cloud session data is fetched using your own logged-in browser session and never leaves your machine except as the final Pulse entries you approve — the same guarantee local capture already gives you. See Privacy & data isolation for the broader picture.

See also

• ⚡ /pulse-capture (Claude Code skill)