v3.41.0

Minor Changes

• a520f7d: Add a fill alias for fast text input without per-keystroke delays.

Patch Changes

• 418ebaa: AI Action no longer runs Playwright code in JavaScript steps; the usePlaywrightPage option has been removed
• 4b575a6: Show a clear, actionable error when a --custom-headers value isn't in HEADER=VALUE form (for example when test paths are passed after the flag) instead of exiting unexpectedly.
• 152b3c8: Object and array env vars referenced in templates now render as JSON instead of [object Object]
• 4b575a6: Show clear guidance when browser installation fails because the browser download directory is not writable
• 4b575a6: Prevent the CLI from crashing when a run artifact file cannot be written
• 4ae0b11: AI Action failures now say when the agent hit the step budget
• 332a64b: Update bundled dependencies to address security advisories
• 4b575a6: Test runs no longer fail to start when organization settings cannot be loaded from Momentic
• 4b575a6: Skip noisy error logs when collecting git metadata in a repository that has no commits yet
• 4b575a6: Show a clear error message when browsers are not available for your operating system, with a workaround, instead of a raw installation failure
• a81fa26: Improve performance monitoring for unexpectedly slow browser test steps.
• 4b575a6: Show a clear, actionable message when your Momentic API key is invalid or expired instead of failing with an unexpected error.
• c37bdda: AI Action steps can now look up the environment variables available to a run, and see the goal you wrote alongside the version with {{env.X}} references filled in
• 07f94cd: Fix bug bash results double-counting a finding that was reported more than once
• 2fed62b: Sign-in automatically issues a new code if you don't finish signing in before the first one expires

v3.40.0

Minor Changes

• 8a7bbe7: AI Action steps are now aware of the surrounding test steps, helping them choose end states that set up the steps that come next

Patch Changes

• 956c3fc: Remove Raindrop analytics integration
• fdff230: Show a clear "page not found" screen when opening an out-of-date link in the local app

v3.39.0

Minor Changes

• f089da6: AI Action can now run Playwright code in JavaScript steps via the new usePlaywrightPage option, letting the agent inspect the live page while exploring and building steps
• f7af475: Allow AI triage to skip recovered tests that do not need permanent healing.

Patch Changes

• e3e23ad: Show a clear, actionable message when your Momentic API key is invalid or unauthorized instead of failing with an unexpected error.
• 1a9ea1a: Keep pending heals non-blocking until healing conclusively leaves them unattempted or fails them.
• 51c6130: Improve reliability of uploading large exploration results
• 7e180d6: Explore findings now distinguish product bugs from UX issues, and exploration timelines reliably show every sub-agent lane.

v3.38.0

Minor Changes

• 1f00b19: Rebase iterative auto-heal pull requests and review changed Momentic files

Patch Changes

• 315c5c3: Fix a rare failure that could corrupt run result archives when they grew past the zip size limit
• 7f9b9fc: Fix ambiguous tab selection in the local editor when multiple tabs share the same URL
• c849a74: Element checks stop re-running AI element location when the page has not changed since the last time it could not find the element.
• 8cfad70: Improve cached element validation to avoid stale cache hits when an element's text content was truncated
• d731fe9: Smart waiting no longer re-asks the AI about a page that has not changed since the last time it answered "not ready", and its model chain now demotes providers that are failing or responding slowly instead of retrying them first on every request.
• 3281b0a: Fix a rare crash when authoring or healing a test with a malformed step
• f6ce0b9: Allow AI Select runs to target test paths, consider pull or merge request descriptions, append custom selection instructions with repeatable --prompt flags, and honor their configured test budget.
• 0f0b51c: Smart waiting now paces by the minimum gap between checks rather than sleeping a fixed interval after each one, so a slow check adds no extra delay.
• d0a93c9: Raise network-log circuit-breaker thresholds for remote browsers and auto-resume capture 30s after it trips

v3.37.0

Minor Changes

• 4e2a001: Give the Explore agent live network-log inspection tools (search_network_requests, get_network_request) as supplemental bug confirmation

Patch Changes

• 812ddd5: Stop rewriting test and module files with formatting-only changes; saving a test or module whose content doesn't change no longer produces a diff.
• 9f73635: Result classification can now inspect browser console and network logs when diagnosing failed web runs.

v3.36.0

Minor Changes

• 52f2168: Fail heal invocations before quarantining eligible non-bug tests that were not repaired or attempted.

Patch Changes

• 3a82b5a: Cancel in-progress browser steps when an AI agent run stops.
• 9a9772e: Keep failed heal attempts for quarantined tests non-blocking unless quarantine is explicitly ignored.

v3.35.0

Minor Changes

• f51d01d: Add momentic_submit_result_classification MCP tool for recording the result classification on a specific run.
• ae3e0f0: Add support for agent skills under .momentic/skills for tuning agents such as explore, result-classification, and triage agents.

Patch Changes

• 4b3d4c7: Discover agent skills recursively within .momentic/skills.
• e53d24c: Change how model streams recover when they fail to call tools correctly.
• 36e0fd1: Improve git context in result classification.
• 106bafd: Teach triage agents how to read file-backed evidence in nested Momentic projects.

v3.34.0

Minor Changes

• 275a8e1: Stop AI Triage at its wall-clock timeout while preserving and uploading partial results.

Patch Changes

• 360fc53: Improve the way manual classification overrides feed into future classifications.
• b7671cd: Improve --share-diagnostics trace clarity.

v3.33.0

Minor Changes

• 7d0ee56: Add run assertions: plain-English checks verified after each test run by an agent that watches the recorded video.
• 8072ec8: Allow AI Triage healing agents to inspect code and use file-backed evidence.

Patch Changes

• 2339b11: Make triage repairs model each required user action explicitly instead of relying on ambiguous retries or side effects.
• cf4bca5: Tune the triage agent's adherence to surrounding authorship styles.
• 22fb8bb: Protect browser responsiveness by stopping excessive console diagnostics and recovering stalled scoped CDP sessions.
• 1ca26f6: Improve tracing for ai explore and triage.

v3.32.1

Patch Changes

• 8fa82d4: Capture step screenshots by default during AI triage.
• a046445: Return web and mobile editor step results faster.
• c97b061: Improve reliability and reduce AI usage during ai heal orchestration.
• 10a0aa8: Prevent agents from making unnecessary tool calls while editing test state.
• 9831058: Stop writing AI action cache data into test files when healing tests