Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Teams

Currents Sessions: one link per agent task

Written November 13, 2026

Currents' Sessions gives every recorded agent task a shareable link. What a session captures, per Currents' own material.

The problem it answers

Agents do the work, but they rarely show it. Currents, a test-reporting platform for Playwright and other frameworks, puts the gap in one line: you get a paragraph of text and have to take the agent's word for it. On November 12, Currents added Sessions to close that gap. When an agent finishes a task, it saves what it captured as a Currents session, and the team gets one link in return.

For a team lead, the link is the artifact that answers the standing question: show me what the agent did. Not its summary of what it did, but the recording itself, in a form both a person and another agent can open.

What a session holds

A session starts with a browser recording, but it is not limited to one. The agent uploads what it captured with three commands: the Playwright trace, the video, screenshots, and page snapshots, along with logs or any other file. Next to the trace and the video, a session can carry the backend log, a HAR file with every network request, the browser console, results as JSON or CSV, and a summary in Markdown. Currents gives the homely example of an agent testing four postcodes after a fix and saving ten files from that single task into one session.

On the session page, each file opens in place. Screenshots and videos play where they sit, Playwright traces open in a trace player, and page snapshots appear as accessibility trees. Two screenshots can sit side by side, two page snapshots can be compared to see what changed, and a single file can be linked on its own.

Two readers for one link

The link comes in two forms. The page itself is for people, and it can be posted in Slack, in Linear, or on the pull request. Add .md to the page link and the same session arrives as Markdown for other agents, with a trace API link for each trace. Currents' worked example runs the full loop: support tags an agent in Slack about a checkout that rejects a postcode, the agent reproduces the fault with Playwright and replies with a session link, and a coding agent then reads the session, finds the failed request in the trace, and fixes the fault without having to reproduce it first. After the fix, the agent records the same steps again as a second session, this time passing, and both links sit on the pull request and the issue.

Agents can also pull new assets out of a recording through the trace API: a filmstrip or an animated GIF of the steps, the frame at the moment of failure, a page snapshot, or the network requests as a HAR file.

Sharing, keys, and lifetime

Limits

The public link is the feature and the exposure at once. Anything inside a session, traces and logs included, is readable by whoever holds the link for as long as it lasts, so a session is the wrong place for recordings that contain secrets or customer data. Sessions also lean on the Playwright world Currents already serves; the capture story is a browser recording plus attached files, not a general screen recording of any agent on any task.

Related playbooks

Sources

Back to: Teams playbooks hub · Also: Coding-agent traces in advisory fixes