Make red mean broken. Fix it in minutes.
Currents separates flaky tests from real regressions, so you can trust your CI again. Every failure comes with the evidence your team, or your agent, needs to fix it.
12 failed
Currents
11 flaky
Quarantined
Still runs, won’t block
1 broken
Fix with AI
Fixed
in 4 min
Every flaky test teaches your team to ignore red
A retry that goes green hides the flake. Do it often enough and nobody reads red anymore. Coding agents learn the same habit: asked for a green build, they add a wait and move on.
Now with . Every guess gets an answer.
Currents tells flaky from broken on every run
- Every execution on record. Test History shows each execution of a test across runs and branches: red failed, purple passed but flaked.
- Broken tests, by name. Project Overview lists the tests that passed, then failed every run since on a branch.
- On the pull request too. The GitHub run comment prints each test's last 30 days, with the base branch beside it.
Keep flakes out of the build, not out of sight
- Out of the build, still on record. A quarantined test still runs and reports its results. With Playwright, its failure doesn't fail the run.
- Your agent writes the rule, in view. Ask it to quarantine a test until Friday. The rule sits in Actions with the rest, and Currents alerts Slack or email before it lapses.
- Or make flakes block the merge. Teams that want it strict set failOnFlakyTests, and any flaky test fails the run.
Red to fixed in minutes, by you or by your agent
10:02. Currents comments on the pull request: checkout › applies a promo code failed, and one other test flaked. It failed 2 of 2 runs on this branch and none on main, so this change broke it.
View trace opens the failed test in Currents, and AI Failure Analysis has read its steps, trace, console and screenshots. Likely cause of the failure: The test clicks a button named Apply. The page now shows Redeem, so the click waits 5 seconds and times out. The fix to try: find the button by its new name, redeem.
Fix with AI hands the failure to Claude Code. It reads the error, the steps and the trace through the Currents MCP server, and finds that 3f9c2e1 renamed the button on purpose. About 1,600 tokens for one failing trace in our example, instead of about 20,000.
Either way, one line changes: the test now looks for Redeem. Your agent re-runs the test to a pass before it pushes. It fixes the test the way the Playwright Best Practices skill teaches.
10:09, seven minutes after the red comment. The run on 8c41d07 reads 1,214 passed, the Currents Tests check passes and the pull request can merge. Same evidence. Same fix.
The right cause, nearly 4× as often
Given the same CI failures, a model found the right root cause nearly 4 times as often with Currents' failure analysis as with the Playwright output and the test code alone.
The alert reaches the owner
Annotate a test or set a mention rule, and the Slack alert mentions its owner, with Fix with AI on it for their agent.
The issue comes filled in
From the failure in Currents, create or link a Jira or Linear issue that carries its error and stack trace.
Let agents fix tests without letting them hide failures
The Playwright Best Practices skill teaches web-first assertions, solid locators and how to find a flake's real cause. It installs from GitHub, no Currents account needed.
npx skills add https://github.com/currents-dev/playwright-best-practices-skillAn agent never gets more than the role of the person who connected it. Connected Applications lists every agent, and removing one ends its access.
An agent's rule sits in Actions with every other rule, where your team sees it and an admin can change or disable it.
A web-first assertion stays green. If your agent only made the wait longer, the test flakes again, and Test History shows it.
Fix the causes that cost the most
- One cause behind many red tests. Error Explorer groups Playwright errors by category, action and target, and counts the tests each one breaks.
- Ranked by what each test costs. Test Explorer sorts by impact, rate times executions, so a test that runs often can outrank a flakier one.
- The CI time red costs you. Project Overview adds up the CI time spent on runs that finished with failures.
Watch your flaky rate fall
- Better or worse at a glance. Six Project Overview tiles show the change since the previous period.
- A year of history. Analytics keep 12 months, enough to show a fix held.
- Reports that come to you. A scheduled summary lands in email or Slack, so nobody has to log in.
Sherpany cut flaky tests from 5% to 0.1%
“Currents gave us the tools to lay a solid foundation for a fast, reliable end-to-end test suite.”
Three steps until red means broken
Minutes from signup to your first recorded run.
Record every run
Start a trial, then add the reporter and your project's keys to CI.
Start a 14-day free trialRead the setup guidePrompt for your agentSet up this repository to report test results to Currents. Read https://docs.currents.dev/getting-started/setup-with-an-agent.md and follow it.Gate the merge
Install the GitHub App and make the Currents check required. Flaky tests are marked and real failures block the merge. GitLab and Bitbucket too.
GitHub App docsPull requestSome checks haven’t completed yetAll checks have passed1 queued1 successful checkCurrents TestsRequiredWaiting to run1,204 passed · 0 flaky · 6m 12sConnect your agent
Add the Currents MCP server and sign in. Your agent reads failures, history and traces.
Connect Cursor, Codex or ChatGPTTerminalclaude mcp add --transport http --scope user currents https://mcp.currents.dev/mcp
