Skip to content

Android agent-device record truncates MP4 evidence at 180s due to adb screenrecord default time limit #615

@nmel-cs

Description

@nmel-cs

Description
While capturing mobile QA evidence for a workspace invite flow on a physical Android device via agent-device record start / record stop, the resulting MP4 was silently capped at 180 seconds even though the driven scenario continued for longer.

What was observed:
Recorded video: https://github.com/user-attachments/assets/98e3eef0-5f94-447f-bbd9-dfae43dee00c
Duration: exactly 180.03s
Gesture telemetry in the same run continues to ~245s — interactions after the 180s mark are not present in the video
No warning or error was surfaced when recording hit the limit; record stop produced a valid-looking MP4

Expected vs. actual behavior

Expected agent-device record captures the full driven session from record start through record stop, including all final verification steps, or explicitly warns when the capture limit is reached.
Actual Recording stops automatically at 180 seconds (ADB default). Later steps (e.g. second Invite member flow, landing on Confirm details) occur on-device but are missing from the MP4 artifact. Validation evidence is incomplete and can be misread as a full end-to-end pass.

Repro steps
prompt:
use agent-device and run the scenario (if u stuck for 30 seconds fail fast)
install the app on connected android device (~/Downloads/adhoc.apk)
launch the app
generate email for our scenario, i.e. random_name_surname+random_number(limit to 99999)@gmail.com
use this email to sign in
members emails can be similar

Tap on the workspace > Overview page
Tap the Invite button
Enter and select an email address
Tap Next > Tap Invite
Tap the Invite member button
Add an email address > Tap Next
Make sure user is landed on the Invite message screen without any crash
FULL VIDEO

Environment:

Field Value
Device Google Pixel 7 (real device)
OS Android 16 (API 36)
App adhoc — ~/***.apk
agent-device 0.15.2
ADB screenrecord v1.4 (default --time-limit 180)
LLM Composer 2.5 Fast

Severity / impact
Severity: Medium
Impact: Final steps may be missing from test artifacts — this directly affects the reliability of validation results. Reviewers or automation may assume the MP4 proves the full flow when it only covers the first 3 minutes. Longer flows (cold start + onboarding + multi-invite) are especially affected on Android.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions