test(commands): cover gesture, settle-baseline, appstate, alert-wait, and JSON-object branches

[thymikee]

Summary

Burns down the five coverage-based CRAP findings in #3422 by covering their uncovered branches through each owning module's public entry point (no new production exports, no test-only DI). 5 test files, one new (interaction/index.test.ts mirrors interaction/index.ts); no production-code changes. Gross diff ~325 lines.

Function	Before	After
cli-grammar/common.ts:227 readJsonObject	30 (0%)	5 (100%)
interaction/index.ts:619 run (gesture dispatch)	35 (25%)	8 (100%)
interaction/runtime/settle.ts:249 sessionBaseline	30 (0%)	5 (100%)
system/output.ts:67 formatAppleAppState	42 (0%)	6 (100%)
system/runtime/system.ts:448 normalizeAlertWaitResult	30 (0%)	5 (100%)

• readJsonObject: accept + all refusal shapes, tested in common.test.ts.
• Gesture run: all seven kinds routed via the family definition's invoke (the real dispatch path), pinning both the client.interactions.* method and the projected options for each kind.
• sessionBaseline: ref-frame / snapshot / empty pre-frame chain through settleAfterInteraction with a point target; a ?? swap mutation fails the ref-frame test.
• formatAppleAppState: runner/session/no-identity renders through systemCliOutputFormatters.appstate.
• normalizeAlertWaitResult: visible / timed-out / no-timing-fields / wrong-kind refusal (typed BackendAlertResult double, asserting COMMAND_FAILED) through device.system.alert.

Closes #3422, part of #3414.

Validation

Tested SHA ffd2f0590: pnpm format and pnpm check:affected --run (under the machine lock) pass. Per-file vitest related + pnpm check:coverage-crap --all over the five owned files: no function at or above CRAP 30 (before: all five listed); CI's Coverage CRAP step confirmed the same at 06d2bb965. Mutation checks: swapping sessionBaseline's ?? order and flipping the alert code both fail their tests. No baseline/gate files touched.

[cubic-dev-ai]

No issues found across 5 files

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

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.17 MB	5.17 MB	0 B
Package (unpacked)	5.17 MB	5.17 MB	0 B
Package (download)	1.55 MB	1.55 MB	-1 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	26.4 ms	26.6 ms	+0.2 ms
CLI --help	79.1 ms	78.6 ms	-0.5 ms

[thymikee]

This PR is ready at 06d2bb9. The diff is test-only, all 18 checks pass, and no open reviewer threads remain. There are no conflicts. I did not run the tests, the mutation checks or the coverage-CRAP gate, and the CRAP before and after numbers are your local measurement. I also did not check that the settle fixtures behave exactly as the assertions assume, so I relied on CI being green for that.

Not blocking: in https://github.com/callstack/agent-device/blob/06d2bb9/src/commands/system/runtime/system.test.ts#L123 the backend double and the wrong-kind alert case return as never, which skips BackendAlertResult typing, so typing those outcomes would let you drop the casts and assert code COMMAND_FAILED with the message. In https://github.com/callstack/agent-device/blob/06d2bb9/src/commands/interaction/index.test.ts#L49 the routing table checks only the method name, so a wrong option mapping for fling, swipe, rotate or transform would pass, and you could add toMatchObject on options for those or leave it to the toXOptions tests in gesture.test.ts. Take or leave either one.

[thymikee]

Both review items addressed in ffd2f0590:

1. system.test.ts: the alert-wait backend double now returns typed BackendAlertResult values (both the projected-outcome array and the wrong-kind outcome), so the as never casts are gone; the wrong-kind refusal asserts code COMMAND_FAILED alongside the message (mutation check: flipping the thrown code to INVALID_ARGS fails the test).
2. index.test.ts: the routing table now pins the projected options for every kind via toMatchObject, including fling, swipe, rotate, and transform — gesture.test.ts only covers the CLI reader and daemon writer for pan/drag, not the client-side toXOptions projection, so these mappings were unpinned before.

Still test-only. pnpm format and pnpm check:affected --run pass on ffd2f0590.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-11 17:29 UTC

[thymikee]

The earlier clean review of 06d2bb9 still holds, and I found no problems in the new tests at ffd2f05. This change is test-only (+0/-0 production lines), and the new cases match the head index.ts projections as I read them. All 18 checks pass and there are no conflicts. I read the diff but did not run the tests on this head, and I did not re-run the mutation check that flips the thrown code to INVALID_ARGS. No review threads are open, so nothing else is needed before merge.