What is the correct way to trigger a screen recording policy from an Architect script step? Flow configured to start capture after transfer node completes. Chrome runs at 124.0.6367.91 on Windows 11. Agent Desktop extension sits at 2.8.1. When script fires, recording policy never activates. Extension console shows manifest v3 background script blocked by default flags.
Storage quota looks completely empty. Codec reports h264 but extension logs say vp9 mismatch right after flow hits recording step. Script just hangs there doing jack all. Mic stays hot but video pipeline never initializes.
Does the config need a flip on the chrome://flags/#extensions-on-chrome-urls toggle or is the flow missing a permission node in JSON? Policy ID matches exactly what sits in Architect workspace. Tried restarting browser three times. Flags reset automatically after each reboot.
Basic question here. Should script step wait for browser event listener before sending start command? Docs say it’s an async trigger but flow times out after twelve seconds. Codec fallback isn’t configured in policy either. Maybe extension won’t handshake with new Chrome security model. chrome.debugger API throws access denied when flow tries to attach to tab. Storage path points to local cache directory. Everything looks standard but pipeline stays dead.
Error: net::ERR_BLOCKED_BY_CLIENT. The Manifest V3 shift breaks the old extension handshake. Chrome 124 drops those legacy messages hard. You’ll want to skip the extension route entirely. Don’t fight the manifest changes. It’s a dead end anyway. Hit the recording endpoint straight from a Data Action instead. Set up a custom action in Architect that fires a POST to /api/v2/screenrecording/sessions. Pass the payload above with the screen-recording:edit scope attached to your token. Terraform’s drift checks usually ignore these dynamic triggers. State files get messy quick. Bypassing the UI hooks keeps things tidy. The endpoint handles agent context validation automatically if the flow already has an active interaction. Push it to prod and watch the session IDs roll in.
The Manifest V3 shift tanks the legacy handshake. Relying on chrome.extension.sendRequest is like running keyword spotting with a noise floor set too low. You’ll get all the garbage and miss the signal. It’s a mess. Switching to a direct Data Action call fixes the precision.
The community suggestion above about hitting the endpoint straight works. Here is the payload structure for the Architect step. It bypasses the extension entirely.
This keeps the recall high for capture events. The error logs don’t flood anymore. Sentiment calibration on the agent side improves when the UI stops throwing red flags. The system won’t flag false positives if the reason code matches the policy. Just ensure the token scope includes analytics:write. The 403s vanish.