fix(desktop): spin the sidebar dot for starting cloud runs

[adamleithp]

Problem

Follow-up nits on top of PostHog/code#4008, which reworked the sidebar row status dots. That PR folded queued into pending, so a cloud task waiting on a sandbox showed the same still yellow dot as a run that had simply never been closed out. Those are not the same fact to a reader. A sandbox being claimed is bounded motion the backend leaves on its own, and it should look like it.

What the dot does

One node changes. A cloud run's first few seconds stop reading as "nothing is happening".

Cloud run, before:

flowchart LR
    A(["send"]) --> Q["queued<br/>● Pending"]
    Q --> S["streaming<br/>⠋ Working"]
    S --> P["in_progress, idle<br/>● Pending"]
    P --> D["PR opened<br/>○ All caught up"]
    classDef phBlue fill:#1d4aff,stroke:#1d4aff,color:#fff;
    classDef phYellow fill:#f9bd2b,stroke:#f9bd2b,color:#000;
    classDef phGray fill:#e5e7eb,stroke:#c7ccd1,color:#000;
    class A phYellow;
    class Q,P phGray;
    class S phBlue;
    class D phGray;

Loading

Cloud run, after:

flowchart LR
    A(["send"]) --> Q["queued<br/>⠋ Starting"]
    Q --> S["streaming<br/>⠋ Working"]
    S --> P["in_progress, idle<br/>● Pending"]
    P --> D["PR opened<br/>○ All caught up"]
    classDef phBlue fill:#1d4aff,stroke:#1d4aff,color:#fff;
    classDef phYellow fill:#f9bd2b,stroke:#f9bd2b,color:#000;
    classDef phGray fill:#e5e7eb,stroke:#c7ccd1,color:#000;
    class A phYellow;
    class Q,S phBlue;
    class P,D phGray;

Loading

Local run, unchanged:

flowchart LR
    A(["send"]) --> Q["queued<br/>● Pending"]
    Q --> S["streaming<br/>⠋ Working"]
    S --> P["never advances<br/>● Pending"]
    classDef phBlue fill:#1d4aff,stroke:#1d4aff,color:#fff;
    classDef phYellow fill:#f9bd2b,stroke:#f9bd2b,color:#000;
    classDef phGray fill:#e5e7eb,stroke:#c7ccd1,color:#000;
    class A phYellow;
    class Q,P phGray;
    class S phBlue;

Loading

⠋ spins, ● is still and solid, ○ is hollow and quiet.

Timings below are from a real cloud run in the dev app, polled once a second from the moment of send:

t	dot	why
0-4s	row not in the list yet	task still being created
5-12s	⠋ Starting	cloud queued, sandbox being claimed. This is the new state
13-14s	⠋ Working	agent streaming, isPromptPending
15s+	● Pending, no work in flight	in_progress with nothing streaming

Eight seconds is long enough that the old still dot read as a click that did not register.

Changes

taskStatusVocabulary.ts:

• Cloud queued now earns the spinner, labelled "Starting". The backend advances that state by itself, so the motion terminates.
• Local queued does not. Nothing writes a terminal status when a local agent goes idle, so a local run can sit at queued for hours after the work is done. It keeps the still dot alongside in_progress runs whose claim outlives the work.
• Comments reworked to say why each case gets the treatment it does, rather than restating the branch.

ChannelItemRow.test.tsx: two cases, the new starting state and its precedence over an existing PR.

Note

The cloud-queued branch returns before the PR check, so a re-queued cloud run on a task that already has a PR spins rather than going quiet. That is deliberate. The PR rule exists because in_progress lies (the workflow holds it while babysitting CI); cloud queued does not lie, and a re-run genuinely is starting. The docblock now says the PR beats a status that merely claims work, not one that is visibly starting, and a test pins it.

The changes originate on the PostHog/code branch ux/task-icons-rework-nits (commit 0edbe86). No PR was ever opened there, and PostHog/code main is frozen after the desktop import, so it is remade here. It only became portable once #76339 resynced products/desktop to fc991d3, which brought PostHog/code#4008 and the file this edits into the monorepo.

How did you test this code?

I (Claude) ran these from products/desktop, on the rebased branch:

• pnpm install --frozen-lockfile, clean
• pnpm --filter @posthog/ui test, 2499 tests across 301 files pass, including both new ChannelItemRow cases
• pnpm --filter @posthog/ui typecheck, 3 errors, all pre-existing. I ran the same command on the base commit and got the identical 3 (CLOUD_STREAM_IDLE_TIMEOUT, MCP_GATEWAY_FLAG, @posthog/harness/extensions/posthog-provider/model-catalog). None are in files this PR touches.

Then I drove the running dev app over CDP with agent-browser and started both a cloud run and a local run in a space, polling the row's accessible name once a second. The dot's accessible name is the exact label string taskDot() returns, so the snapshot reads the real value rather than a rendering of it. The cloud table above is that run. The local run sat at "Pending, no work in flight" from creation onward and never spun, which is the branch this change deliberately excludes.

Raw polling output, cloud run

--- t=4s
    - button "Pending — no work in flight Respond with greeting message Cloud"
--- t=5s
    - button "Starting Say hello Cloud"
--- t=12s
    - button "Starting Say hello Cloud"
--- t=13s
    - button "Working Say hello Cloud"
--- t=15s
    - button "Pending — no work in flight Say hello Cloud"

The two new tests assert that a cloud run at queued gets the spinner while a local run at queued does not, and that the spinner survives an existing PR on the task. No existing test covered the workspace mode split (both used to fall through to the same still dot), and the two PR-precedence cases both used in_progress, so nothing pinned the interaction between a starting run and a PR.

Automatic notifications

• Publish to changelog?
• Alert Sales and Marketing teams?

Docs update

No user-facing docs cover the sidebar dot vocabulary.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Ported by Claude Code (Opus 5) using the repo's /porting-code-prs skill, with /writing-code-comments and /writing-tests for the follow-up rounds. The skill's normal path is gh pr diff into git am --directory=products/desktop/, but there was no source PR here, just a branch with one commit, so the patch came from git format-patch on the fetched branch instead. git am -3 preserved the original authorship.

The interesting part was the base. The skill is explicit that merged PostHog/code changes must arrive through the resync protocol in MIGRATION.md and must never be hand-ported, since that duplicates the diff. PostHog/code#4008 is merged and sat after what was then the pinned SHA, so the file this commit edits was absent from master and the port was blocked. I first built this stacked on the open resync PR, then found #76339 had already merged an hour earlier, so it was rebased straight onto master and the stack thrown away.

A review pass afterwards found the PR-precedence gap called out above, plus a docblock that claimed queued folds into working without qualifying it to cloud. Both are fixed in follow-up commits. The commit message was reworded from nits to conventional-commit format.

[trunk-io]

😎 This pull request was merged.

[trunk-io]

   

View Full Report ↗︎ ⋅ Docs

[greptile-apps]

_{Reviews (1): Last reviewed commit: "test(desktop): pin the starting spinner ..." | Re-trigger Greptile}

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-08-01 14:37 UTC	Run
prod-us	✅ Deployed	2026-08-01 14:51 UTC	Run
prod-eu	✅ Deployed	2026-08-01 14:53 UTC	Run