[richiemcilroy]

Instant mode's worst failures happen in the real app, across process deaths: a crash during start or capture, a start stopped while it is still in flight, a recovery on the next launch. The unit and mock-server tests cover each piece in isolation; nothing ran the whole app through them. This adds a suite that does, and a macOS job that runs it on every pull request touching the Instant path and nightly.

Builds on #3106.

flowchart LR
    T[cargo test real_app --ignored] --> M[MockCap on 127.0.0.1]
    T --> S["Sandbox: store pointing at the mock,<br/>own recordings and logs"]
    S --> A[Built cap-gpui binary]
    A -- "create, presign, segments,<br/>recording-complete, delete" --> M
    T -- "kill -9, relaunch,<br/>CAP_GPUI_AUTO_RECORD / AUTO_RECOVER" --> A
    T --> C["Check the mock's created, deleted and stored objects,<br/>and the sandbox's projects and journal"]

Loading

Cases (apps/desktop-gpui/src/upload/real_app_tests.rs)

• real_app_instant_recording_uploads_and_verifies: a normal recording reaches Verified, the mock holds its confirmation, nothing is deleted and the cleanup journal is empty.
• real_app_crashes_never_delete_a_recording_with_media: kills the app at 8 points, from the server video's creation through the first seconds of capture, noting each recording's folder and media right after its kill, then ages the leftovers and relaunches. A recording that held media keeps its folder, its media and its server video, and every crashed recording that held none in a folder still in progress has its server video deleted.
• real_app_recovered_recording_uploads: kills the app after four finished segments, relaunches with CAP_GPUI_AUTO_RECOVER=1, and requires the recovered recording to verify under the same video id with its segments stored.
• real_app_cancelled_start_deletes_its_video: stops a recording while its start is in flight. The server video it created is deleted and no project or journal entry is left.

They are #[ignore]d because they need a built app and screen recording permission:

CAP_GPUI_REAL_APP_BIN=target/debug/cap-gpui \
  cargo test --bin cap-gpui real_app -- --ignored --test-threads=1

CAP_GPUI_REAL_APP_ARTIFACTS keeps each sandbox (app logs, recordings, journal). If the app does not reach a step in time, its threads are sampled into hang-sample.txt before the test fails, so a hang can be diagnosed from the job's artifacts.

Mock server

MockCap now records the video ids it creates and deletes (created(), deleted()), so a test can tell which recording's video was removed.

CI (.github/workflows/instant-reliability.yml)

• Runs on Blacksmith macOS 15 with the GPUI build cache, for pull requests that touch the Instant upload, recording, cleanup, recovery or segment muxing code, nightly, and on dispatch. A scope job runs it for this repository's own changes, including workflow edits, and for fork pull requests only when they change Instant code.
• Grants Screen Recording to the app and test binaries, wakes the display and adds BlackHole, as verify-recording.yml does (the start chime plays before capture begins).
• Uploads the sandboxes when the job fails.

Confidence Score: 5/5

The changes since the previous review appear safe to merge.

Summary

This revision restores the workflow’s own path trigger and adds a small job that decides whether to run the macOS suite.

• Real-app tests check Instant uploads across crashes, stops, and recovery.
• A macOS job runs real-app Instant checks on relevant changes.

All supplied previous threads were unnumbered and manually resolved by richiemcilroy without explanatory replies. The current code counts unfinished media, retries missed cancellations, stops child apps on drop, and removes the accessor comments.

_{Reviews (5) · Last reviewed commit: "ci: run the Instant reliability job on i..." · Reviewed by Greptile}

[richiemcilroy]

hey @greptileai, please re-review the PR

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
cap-status	Ready	Preview	Oct 24, 2026 3:51pm UTC

[richiemcilroy]

hey @greptileai, please re-review the PR

[superagent-security]

Superagent found 1 security concern(s).

[capy-ai]

Capy found 1 potential issue (1 medium).

View 1 other finding in Capy.

[richiemcilroy]

hey @greptileai, please re-review the PR

[richiemcilroy]

hey @greptileai, please re-review the PR

[capy-ai]

Capy found no issues.

View 3 other findings in Capy.