--- title: "Sessions: record every agent task" description: "Your agent saves its browser recording, or the files of any task, as a Currents session. You get one link: a page for people and a Markdown version for other agents. Post it in Slack, Linear or on the pull request." canonical: "https://currents.dev/changelog/agent-sessions" updated: "2026-11-12" --- # Sessions: record every agent task 2026-11-12, New (tracks: platform, cli, api) Your agent saves its browser recording, or the files of any task, as a Currents session. You get one link: a page for people and a Markdown version for other agents. Post it in Slack, Linear or on the pull request. --- When an agent says it reproduced a bug or fixed it, you can now see what it did. The agent saves its browser recording as a Currents session, and you get one link: a page people open in the browser, and a Markdown version other agents can read. Agents do the work, but they rarely show it. You get a paragraph of text and have to take their word for it. With sessions, the agent uploads what it captured to Currents with three commands: the Playwright trace, the video, screenshots, page snapshots, logs or any other file. Anyone with the link can open it, no Currents login needed. ## Capture bug reproductions Agents can capture bug reproduction evidence for replays and investigations. When an agent reproduces a bug with Playwright, it records a session: the trace, the video, the screenshot of the error and the page snapshot. Anyone can replay what happened and start the investigation from the recording, not from a written description. ![Slack: support reports that a customer can't check out with postcode SW1A 1AA and tags Claude. Claude replies: On it. Reproducing in a browser and recording a session.](/img/changelog/agent-sessions-1.jpg "Support tags the agent in the Slack thread") For example, support tags the agent in a Slack thread about a checkout that rejects a postcode. With one message, support produces the evidence instead of a bug report in words, on one page anyone in the thread can open. ## Point agents to the recording Coding agents can work from a session instead of a description. They read the session as Markdown, find the failed request, and get its request and response from the trace through the [trace API](https://docs.currents.dev/ai/reference/trace-api). The agent doesn't have to reproduce the bug again, and it starts from what actually happened. For example, the Linear issue links the support session. The developer's agent reads it and finds the cause: the postcode check rejected postcodes with a space. It fixes the check, runs the same steps again and records them as a second session, this time passed. ## Share the session where the work happens A session link goes wherever your team already works. The agent posts the before and after on the pull request and on the Linear issue, with the screenshots inline and both sessions one click away. The reviewer sees the fix work without checking out the branch, and whoever reads the issue later sees what was broken and what changed. Agents can also generate new assets from a trace when they need them, through the [trace API](https://docs.currents.dev/ai/reference/trace-api): a filmstrip or an animated GIF of the steps, the frame at the moment of the failure, a page snapshot, or the network requests as a HAR file. They post exactly the picture that explains the change, made from the recording itself. ![A pull request comment with the before and after screenshots and links to both sessions](/img/changelog/agent-sessions-2.jpg "The before and after on the pull request") ![A Linear comment from Claude with the before and after screenshots and the two session links](/img/changelog/agent-sessions-3.jpg "The same evidence on the Linear issue") ## Every file the agent collects, on one page A session isn't limited to browser recordings. Next to the trace and the video, the agent can attach the backend log, a HAR file with every network request, the browser console, its results as JSON or CSV, and a summary in Markdown. Each file opens in place, so whoever gets the link sees the whole picture without asking the agent for more. After the fix, an agent checked four UK postcodes on the fix branch and saved everything it collected in one session: ![A passed session with ten files: a screenshot, a video, a Playwright trace, a page snapshot, server.log, network.har, console.log, results.json, results.csv and summary.md. network.har is open and lists 16 requests, one of them a 422.](/img/changelog/agent-sessions-4.jpg "Ten files from one check, with the network requests open") ## What's on the session page Each file opens in place: screenshots, videos, Playwright traces in a trace player, page snapshots as accessibility trees, logs, JSON, CSV and Markdown. You can put two screenshots side by side, see what changed between two page snapshots, and copy a link to a single file to post it on its own. Add `.md` to the page link and you get the Markdown version for agents, with a trace API link for each trace. ## Get started Three commands, run by the agent or by you: ```bash npx currents session start --project-id --title "Postcode rejected" --status failed npx currents session attach repro.webm screenshot.png .playwright-cli/traces npx currents session share ``` To have your agent do this on its own, give it a Currents skill. Agents connected to the Currents MCP server already have one, and `npx currents skill --install` adds it for agents with a shell. Sessions work with [Playwright MCP](https://docs.currents.dev/ai/agent-evidence/bug-and-fix/playwright-mcp) and [Playwright CLI](https://docs.currents.dev/ai/agent-evidence/bug-and-fix/playwright-cli) recordings, and with any file. The record key your CI already has can create and share sessions. A Read & Write API key can also list and delete them. A session is not a test run, so it doesn't show up in your runs, analytics or flakiness. Share links are public and last 1, 3 or 7 days. The files stay until you delete the session. - [Agent evidence](https://docs.currents.dev/ai/agent-evidence) - [Post evidence to GitHub, Slack and Linear](https://docs.currents.dev/ai/agentic-artifacts/post-evidence) - [currents session](https://docs.currents.dev/resources/reporters/currents-cmd/currents-session)