refactor(daemon): relocate the interaction touch tests by what they assert by thymikee · Pull Request #3672 · callstack/agent-device

thymikee · 2026-10-15T13:53:49Z

Summary

Batch interaction-touch of the daemon test redesign (no test exports of any kind). The 16 root tests (plus the interaction-touch-runner.ts helper and the root-level interaction-touch-android-readiness-spawns sibling) that drove handleInteractionCommands with the real command surface are placed by what they assert. Daemon halves (a) now run against package fakes; assertions about what the real src/commands/interaction/** code does (c) moved to tests beside that code. snapshot-interactor-capture mocks are gone: captures go through installInteractorSnapshotResolution() (the InteractorResolution seam). Stacked on #3670.

New package fixtures, all unexported: interactor-resolution-fixture.ts, touch-command-surface-fake.ts, touch-interaction-runner.ts.

a / c split and test-count parity

Before = exact count at the old path (vitest JSON reporter); after = run count per destination (all under packages/daemon/src/execution/ unless noted).

Source (before)	Daemon half (a)	Command half (c), `src/commands/interaction/runtime/`
`interaction-touch` (12)	`interaction/internal/interaction-touch-routing.test.ts` (12)	none
`interaction-touch-press-admission` (13)	`.../interaction-touch-press-admission.test.ts` (12)	`interactions-preresolved-target.test.ts` (the #1654 provenance-mismatch case)
`interaction-touch-press` (9)	`.../interaction-touch-press.test.ts` (5)	`interactions-target-promotion.test.ts` (4), `interactions-preresolved-target.test.ts` (5, shared)
`interaction-touch-fill` (6)	`.../interaction-touch-fill.test.ts` (4; +1 dispatch-and-record)	`interactions-ref-frame-evidence.test.ts` (off-screen and invalid bounds), `interactions-target-promotion.test.ts` (editable node)
`interaction-touch-runtime` (7)	`.../interaction-touch-runtime.test.ts` (3)	`interactions-ref-frame-evidence.test.ts` (5, shared); the missing-rect refusal is already pinned by `ref-target-resolution.test.ts`
`interaction-touch-response` (6)	`.../interaction-touch-response-evidence.test.ts` (6)	digest/changed-from-before stays in `interactions.test.ts`
`interaction-touch-payload` (7)	`.../interaction-touch-payload.test.ts` (7)	none
`interaction-touch-prepare` (5)	`.../interaction-touch-prepare.test.ts` (5)	none
`interaction-touch-android-freshness` (4)	`.../interaction-touch-android-freshness.test.ts` (4)	none
`interaction-touch-android-readiness` (7)	`.../interaction-touch-android-readiness.test.ts` (7)	none
root `interaction-touch-android-readiness-spawns` (1)	`.../interaction-touch-android-readiness-spawns.test.ts` (1; real `createAndroidObservationAdapter` over a counting `runAdb` host port: six `dumpsys` calls for three taps)	none
`interaction-touch-direct-ios-eligibility` (4)	`.../interaction-touch-direct-ios-eligibility.test.ts` (4)	none
`interaction-touch-direct-ios` (6)	`.../interaction-touch-direct-ios.test.ts` (6, incl. the maestro-direct disclosure row loop)	none
`interaction-ios-tap-outcome` (22)	`.../interaction-ios-tap-outcome.test.ts` (22)	none
`interaction-gesture-drag` (2)	`.../interaction-gesture-drag.test.ts` (2)	drag endpoint disclosure already pinned by `gestures.test.ts`
`interaction-get` (16)	`execution/selector-runtime-get-reads.test.ts` (9), `execution/selector-runtime-is-reads.test.ts` (5)	`selector-is-visibility.test.ts` (2), `selector-read-disambiguation.test.ts` (1)

127 tests before; 131 after (the added ones are the fill dispatch-and-record case, the disambiguation case, and splits). The interaction-touch.test.ts original was deleted with its package copy (12 = 12); nothing was dropped without a named owner.

Selection

All destinations match packages/*/src/**/*.test.ts or src/**/*.test.ts, so unit-core selects them (pnpm test:unit, test:coverage); the sources were already in unit-core. check:affected --run runs the vitest related leg over them.

`vi.mock` (each reads what its mock provided or recorded)

interaction-touch-android-readiness.test.ts, platform-android/mechanics: the mocked getAndroidAppState result is read by "tap likely escaped the app", "foregrounded com.android.settings" and the permission-dialog warning.
same file, ../../android-system-dialog.ts: its spy drives "a ref action aborts with the shared ref_frame_expired rejection after Android dialog recovery mutates" (details.reason === 'ref_frame_expired', mockTapPoint not called).
interaction-touch-payload.test.ts, platform-android/mechanics: getAndroidScreenSize is read by toHaveBeenCalledTimes(1), not.toHaveBeenCalled() and the 1344×2992 reference-size assertions.
interaction-touch-android-freshness.test.ts, platform-android/mechanics: a benign default that keeps the readiness check from spawning adb; the response.ok assertions depend on it. No assertion reads its recordings.
selector-runtime-is-reads.test.ts, platform-android/mechanics: mockGetAndroidAppState is read by the permission-dialog blocker test (foregroundPackage in details).
selector-runtime-get-reads.test.ts, platform-apple/runner/operations: a negative guard. expect(mockRunAppleRunnerCommand).not.toHaveBeenCalled() and the querySelector not.toHaveBeenCalledWith fail if the daemon starts reaching the runner for a get again, instead of silently running it. It passes today because the daemon no longer queries the runner.

Subpath exports

Package	Before (#3670)	After
`daemon`	85	81

Deleted: __tests__/ios-comparison-key-fixture, __tests__/replay-runtime/replay-command-fixture, interaction/internal/interaction-ios-tap-outcome, interaction/internal/interaction-touch-android-readiness. No other package changed; no new subpath, barrel or export *.

Notes

The touch fake's target resolution is a deliberate minimum (points resolve to themselves; a @ref is the centre of its node in refFrameSnapshot/snapshot; a selector is the first matching node of one fresh capture; preresolvedTarget is honored). It has no promotion, occlusion, off-screen, bounds, label-recovery or verify-evidence logic. Those are asserted as (c) tests against the real runtime, so its default answers do not stand in for command behavior.
Real-surface composition legs removed from the daemon tests are covered by test/integration/interaction-contract/* and provider-scenarios/press-selector-resolution.test.ts; I have not confirmed those assert the same cases one for one.
One-line path repoints for registries that named moved tests: packages/contracts/src/dispatch-disclosure.fixtures.ts (maestro-direct. owner) and packages/test-kit/src/boundary-fault-matrix.ts.
(c) tests are additive files in src/commands/interaction/runtime/, outside packages/daemon; say if any belongs to another worker.

Validation

check:quick, check:layering, check:fallow and check:gate-manifest pass on this head; check:affected --run passes (including test-file-size-ratchet and the eager-closure suite).

🤖 Generated with Claude Code

cubic-dev-ai

All reported issues were addressed across 33 files

_{Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic}

thymikee · 2026-10-15T15:41:06Z

Thanks for the careful move. At 8bedf50 the new touch-interaction-runner.ts re-declares contextFromFlags (https://github.com/callstack/agent-device/blob/8bedf50/packages/daemon/src/__tests__/support/touch-interaction-runner.ts#L8), which already exists as an export in https://github.com/callstack/agent-device/blob/8bedf50/packages/daemon/src/execution/interaction/internal/__tests__/interaction-touch-fixtures.ts#L65. The deleted runner imported that export, and the new one already imports the sibling fixture, so nothing blocks reuse. With two copies, a new request flag added to one will make runInteraction and the sibling suites diverge without any failure. Please keep one projection: import it from interaction-touch-fixtures.ts, or move it to support/ and have both import it.

Not blocking, and fine to take or leave: some new test names do not match the module beside them (interaction-touch-routing has no sibling module, interaction-touch-direct-ios-eligibility sits away from its module, and interaction-touch-android-readiness-spawns now counts runAdb calls and spawns nothing), and several daemon-half assertions read values the touch-command-surface fake invents (the always-pass is, the fixed READ_CAPTURE_OPTIONS, the made-up verify evidence), so the test headers should say which are fake-supplied and covered by the real-surface tests, and the as never casts could use the typed return shapes.

Of the open inline threads, none still apply: all nine fixed-at-head or outdated ones can be resolved (#3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment), #3672 (comment)), and the fill-contrast thread is benign because the stale-ref case asserts the fill point is not called (#3672 (comment)).

CI has not run yet: all 13 remaining checks are queued, with no failures. I did not run the tests myself. Before merge, please remove the duplicate contextFromFlags and wait for Typecheck, Repo Guards, Coverage and Integration to pass on the new head.

github-actions · 2026-10-15T15:56:02Z

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.24 MB	5.24 MB	0 B
Package (unpacked)	5.24 MB	5.24 MB	0 B
Package (download)	1.57 MB	1.57 MB	-3 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.7 ms	28.2 ms	+0.5 ms
CLI --help	78.8 ms	80.7 ms	+1.9 ms

…mand surface fake The interactor-resolution fixture replaces the snapshot-interactor-capture module mock: captures reach one spy through the installed InteractorResolution. The touch surface fake dispatches resolved targets through the runtime config's backend and reads the session through its store, so daemon suites pin admission, bound-operation wiring, response shaping, and capture wiring without the real command family. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…nteraction handler interaction-touch.test.ts is (a) daemon logic: router claims press/click/longpress/hover/fill, single inspect/bind, macOS surface refusals, hover capability refusal. All 12 tests move to execution/interaction/internal/interaction-touch-routing.test.ts on the touch surface fake. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…t each asserts interaction-touch-press-admission.test.ts: (a) 12 tests move to the daemon package on the touch surface fake; the #1654 provenance-mismatch test is (c) and moves to src/commands/interaction/runtime/interactions-preresolved-target.test.ts. interaction-touch-press.test.ts: (a) secondary-button projection, the find pre-resolved wiring (adopt, control, no lookup) and read-after-press capture stay in the daemon package; the macOS menubar wrapper promotion, hittable-ancestor promotion, full-screen non-promotion (interactions- target-promotion.test.ts) and the pre-resolved adopt/control/no-lookup/shared-guard cases (interactions-preresolved-target.test.ts) are (c) and run against the real command runtime. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…asserts interaction-touch-fill.test.ts: (a) expired-frame rejection, pinned stale/current ref, and the find-internal fill without a stale-ref warning move to the daemon package (plus one new test that the resolved point, typing delay and recorded ref reach the bound fill). The off-screen refusal, invalid-bounds refusal and editable-node retention are (c): interactions-ref-frame-evidence.test.ts and interactions-target-promotion.test.ts. interaction-touch-runtime.test.ts: (a) the resolved-point tap/record test and the #1542 rescue hook wiring (hook present iff the owner advertises querySelector) move to the daemon package; the invalid-bounds, off-screen refusal and in-frame label recovery cases are (c) in interactions-ref-frame-evidence.test.ts (the missing-rect refusal is already ref-target-resolution.test.ts). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ide their modules interaction-touch-response.test.ts (6), interaction-touch-payload.test.ts (7) and interaction-touch-prepare.test.ts (5) are (a) daemon logic: evidence rides the response allowlist only when --verify ran, recording/visualization payloads and reference frames, and the settle observation capture the request admitted. They move to execution/interaction/internal/ on the touch surface fake, the interactor-resolution fixture and daemonPlatformServicesFixture. The --verify evidence computation itself stays covered by the real-runtime tests in src/commands/interaction/runtime/interactions.test.ts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ide their modules interaction-touch-android-freshness.test.ts (4) and interaction-touch-android-readiness.test.ts (7) are (a) daemon logic: the freshness refresh may not retarget an admitted ref, the escape and permission-prompt outcomes, and recovery aborting an admitted ref. They move to execution/interaction/internal/. The readiness suite now mocks the exact android-system-dialog module the handler imports. interaction-touch-android-readiness-spawns.test.ts is (a): the two-checks-per-tap dumpsys budget is pinned through the real Android observation adapter over a counting runAdb host port instead of the root adb provider and host binding. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…press handler interaction-touch-direct-ios.test.ts (6) and interaction-touch-direct-ios-eligibility.test.ts (4) are (a) daemon logic: the Maestro-only direct selector route, its error passthrough, the fused ref-frame seam, and the maestro-direct dispatch-disclosure rows. They move to execution/interaction/internal/. The dispatch-disclosure driver owner and the boundary-fault-matrix evidence path follow the moved driver file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ir modules interaction-ios-tap-outcome.test.ts (22) and interaction-gesture-drag.test.ts (2) are (a) daemon logic: same-scope corroboration of a failed iOS tap, the recorded drag with both endpoint identities, one inspect/bind per drag, and frame expiry. They move to execution/interaction/internal/ on the touch surface fake (which now answers a target-authored drag through one backend plan). The drag endpoint resolution and disclosure stay covered by src/commands/interaction/runtime/gestures.test.ts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@ref

interaction-get.test.ts: (a) the 9 get tests (bound read expansion, no live read, retired direct selector, frame-tree @ref reads and their warnings) move to execution/selector-runtime-get-reads .test.ts and the 5 is tests that pin capture wiring and the Android foreground blocker move to execution/selector-runtime-is-reads.test.ts. The two predicate-semantics tests (is visible inheriting ancestor viewport visibility, is visible failing outside the viewport) are (c) in src/commands/interaction/runtime/selector-is-visibility.test.ts, and the same-label disambiguation read is (c) in selector-read-disambiguation.test.ts. Nothing imports the root interaction-touch-runner helper any more, so it is deleted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…h fake's drag and ref routes follow production Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…nd say which fake values are fake-supplied Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

thymikee · 2026-10-15T16:00:53Z

Fixed in f311725. touch-interaction-runner.ts imports the one contextFromFlags from interaction-touch-fixtures.ts instead of re-declaring it. The only other copy, in interaction-ios-tap-outcome.test.ts, is a deliberate superset (it adds snapshotPreferredBackend for the corroboration capture), so I left it. The not-blocking asks: the header of touch-command-surface-fake.ts now lists the values the fake supplies itself (is always passes with one match, reads capture with READ_CAPTURE_OPTIONS, --verify evidence is a fake-digest-<n> placeholder) and names src/commands/interaction/runtime/ as where the real values are asserted. I took no action on the test names or the as never casts; both are cosmetic and I'd rather not widen this PR.

cubic-dev-ai

1 issue found across 2 files (changes from recent commits).

_{Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic}

Prompt for AI agents (unresolved issues)


Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/daemon/src/__tests__/support/touch-command-surface-fake.ts">

<violation number="1" location="packages/daemon/src/__tests__/support/touch-command-surface-fake.ts:45">
P3: This overstates capture behavior: `get` on a point or `@ref` does not capture a snapshot. Specify that selector reads use `READ_CAPTURE_OPTIONS` so the documentation does not imply capture wiring for every read.</violation>
</file>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

thymikee · 2026-10-15T16:49:09Z

This PR is ready at 75f4e4c. The two points from the earlier review (#3672 (comment)) are fixed: the shared runner import and the fake's header comments now match the code. Not blocking: packages/daemon/src/__tests__/support/touch-interaction-runner.ts line 3 imports its projection from a feature-local __tests__/ fixture, which points the dependency the wrong way, and if a third consumer appears outside execution/interaction, moving contextFromFlags into __tests__/support would fix it, but you can take or leave this.

The one earlier thread on the fake's header (#3672 (comment)) no longer applies. The header now says that only selector reads (get, is) capture, once, and that point and @ref reads capture nothing, which matches the fake. You can resolve it.

I did not run the test suites. This is a test relocation, so the moved assertions are the evidence. I did not re-diff the full 33-file move, because the earlier review covered it and this change touches only the runner import and two header comments. CI has not run on this head yet, and all checks are queued with none failed. Before merge, CI needs to go green on 75f4e4c, and #3670 needs to land first so this PR's base is main.

github-actions · 2026-10-15T17:13:35Z

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-15 17:13 UTC

cubic-dev-ai Bot reviewed Oct 15, 2026

View reviewed changes

thymikee force-pushed the refactor/daemon-interaction-touch-tests branch from 44dcc5d to cff2d79 Compare October 15, 2026 14:22

thymikee force-pushed the refactor/daemon-interaction-touch-tests branch from cff2d79 to 5b442e7 Compare October 15, 2026 14:35

thymikee force-pushed the refactor/daemon-interaction-touch-tests branch from 5b442e7 to 8bedf50 Compare October 15, 2026 14:36

thymikee added this pull request to stack #3681 October 15, 2026 15:46

thymikee and others added 12 commits October 15, 2026 17:59

test(daemon): delete the touch routing test the package copy replaces

6207002

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

test: assert typed refusal details, count captures, and make the touc…

41e5405

…h fake's drag and ref routes follow production Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

test(daemon): reuse the shared contextFromFlags in the touch runner a…

f311725

…nd say which fake values are fake-supplied Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

thymikee force-pushed the refactor/daemon-interaction-touch-tests branch from 8bedf50 to f311725 Compare October 15, 2026 16:00

cubic-dev-ai Bot reviewed Oct 15, 2026

View reviewed changes

Comment thread packages/daemon/src/__tests__/support/touch-command-surface-fake.ts Outdated

docs(test): say that only selector reads capture in the touch fake

75f4e4c

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

thymikee added the ready-for-human Valid work that needs human implementation, judgment, or maintainer merge label Oct 15, 2026

thymikee merged commit 9c06cd2 into main Oct 15, 2026
20 checks passed

thymikee deleted the refactor/daemon-interaction-touch-tests branch October 15, 2026 16:50

Navigation Menu

Repository navigation

refactor(daemon): relocate the interaction touch tests by what they assert - #3672

refactor(daemon): relocate the interaction touch tests by what they assert#3672
thymikee merged 13 commits into
refactor/daemon-session-lifecycle-testsfrom
refactor/daemon-interaction-touch-tests

thymikee commented Oct 15, 2026 •

edited by cubic-dev-ai Bot

Loading

Uh oh!

cubic-dev-ai Bot left a comment •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026 •

edited

Loading

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

cubic-dev-ai Bot left a comment •

edited

Loading

Uh oh!

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

Uh oh!

github-actions Bot commented Oct 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

thymikee commented Oct 15, 2026 • edited by cubic-dev-ai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

a / c split and test-count parity

Selection

vi.mock (each reads what its mock provided or recorded)

Subpath exports

Notes

Validation

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Size Report

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

thymikee commented Oct 15, 2026

Uh oh!

Uh oh!

github-actions Bot commented Oct 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

thymikee commented Oct 15, 2026 •

edited by cubic-dev-ai Bot

Loading

`vi.mock` (each reads what its mock provided or recorded)

cubic-dev-ai Bot left a comment •

edited

Loading

github-actions Bot commented Oct 15, 2026 •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading