🧠 Docket Technical Changelog

May 2026

Universal modules. One module is reusable across every browser viewport, iOS device, and Android device. The viewport_width, viewport_height, browser_zoom, device_type, and device_name fields have been removed from the module API.
Per-config step caches. Recorded and self-healed steps stay sharp on every device / viewport / zoom they’ve been run against. list-modules adds a config_key query param to scope what’s returned.
New Start Screen launcher. The Start Screen button now opens a device picker grouped by platform with per-resolution entries and inline mobile app selection. Defaults follow the test you’re viewing.
Screen Out of Sync overlay. When the running worker drifts from the test’s config, the stream shows an overlay listing the specific drifts and a single Restart Screen button. Run and Record are disabled until they match.

Optimized AWS + Porter cluster configuration, cutting average browser-worker CPU/memory usage by ~22%.
Tuned pre-check confidence threshold and CPU allocation logic per site for more stable agent runs.
Added automatic weekly dependency vulnerability scans in CI/CD pipeline.
Improved S3 object storage configuration with versioning + Object Lock for data integrity.
Refactored internal monitoring to surface agent latency, memory, and timeout metrics in dashboard.

Enhanced self-healing agent behavior for broken selectors, delayed elements, and network retries.
Introduced usage analytics pipeline capturing test latency, pass/fail counts, and error codes.
Benchmarked and tuned Playwright concurrency profiles across different CPU classes to reduce cold-start time.

Deployed full AWS + Porter environment with staging + production clusters and load-balanced worker pool.
Launched new vision model for Docket
Introduced system health dashboard showing session errors, retries, and agent restarts.
Improved Docker image build flow — reduced image size and deployment time by >30%.