Skip to content
Agents tracked: 258 Downloads (7d): 219M up 6.1% GitHub stars: 5.5M VS Code installs: 148M Releases (7d): 293 Agent status: 2 with issues Updated Oct 7, 2026
AgentGid

TesterArmy e2e changelog: what's new each month

Every stable TesterArmy e2e release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 1 month covered; the current month updates daily.

October 2026 · so far

5 releases: 0.15.1 → 0.18.0

October brought improvements to model support, MCP integration, and replay reliability, along with expanded secret redaction and a Node.js version requirement update. The month saw better handling of edge cases in agent steps and reduced dependency footprint.

Action needed
  • Node.js 22.22.3 or newer on Node.js 22, or 24.8.0 or newer required (was more permissive).
  • Interrupted tests are no longer counted as failed; report.json now has run.summary.interrupted.
  • EnginePrepareInfo.headed is now required alongside the headed value already sent on init.

Highlights

  • DeepSeek V4 models and Claude now work reliably in agent steps with improved provider compatibility.
  • Sign in with OpenCode Console to use OpenCode Zen and OpenCode Go models.
  • MCP sessions are now headless by default with per-session control.
  • Installation is faster and lighter, pulling 29 packages instead of 117.
  • Secrets are now redacted in more places including test titles and text downloads.

New

  • e2e mcp can now open sessions in projects without the optional ai package.
  • screen.swipe() accepts an optional duration parameter in milliseconds.
  • Locator matchers now honor Playwright's { checked: false }, { enabled: false }, { visible: false }, and { attached: false } options.
  • e2e models github-copilot marks which models are unavailable based on plan limits.
  • Device flow login for OpenCode Console integration.
  • Anonymous telemetry for e2e mcp and e2e explore sessions.

Improved

  • Model failures no longer evict replay cache entries when no model answers.
  • The bundled e2e init shows current config shape and setup examples.
  • npm package pages now include repository README and searchable keywords.
  • Documentation pages are now shipped offline in node_modules/e2e/docs.
  • Replay cache no longer passes steps whose effects did not occur.
  • Failed test attempts on unresponsive apps no longer wait indefinitely for failure evidence.
  • expect(value).toBeDefined() now correctly passes on null values like Playwright and Jest.
  • Forced interrupts (second Ctrl-C or CI cancel) now write junit.xml and summary.md.

Fixed

  • DeepSeek's thinking mode no longer causes MODELPROVIDERFAILED on first agent turn.
  • device.installApp() no longer incorrectly pins file paths as app identifiers on Android.
  • Recorded steps that clear fields with empty values now replay correctly instead of failing with REPLAYSTALE.
  • Recorded steps that remove controls replaced by status text now work in replay.
  • Run-level errors in reports are now properly redacted with configured secrets.
  • Claude and similar models that downgrade forced tool choice are now handled correctly.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. TesterArmy e2e: pricing, features and alternatives · All changelogs