[thymikee]

Summary

Follow up on the review of #3849, which has merged:

• Parse complete, correlated touch replies in the protocol owner. Preserve error ordering, helper retirement, and ambiguous-dispatch handling.
• Build shell-fill unconfirmed evidence once. Share warning emission with the other fill paths.

Two commits, five files: production +25/−20, existing test +3/−5. The existing parser test calls the renamed entrypoint; no new tests or cases were added.

Validation

Head 15f17c47ddf81e842c81e348a0b4f5549bd9a06e, base f8c60ca188f62e110bdc1d3e6225b134960886c0:

• pnpm check:affected --run --head 15f17c47dd: passed, 2,316 tests. Receipt .tmp/validation/run-2UkGNn/receipt.json: exactCommit=true, checkoutChanged=false. CLI/native helper builds passed. Claude review: no remaining findings.
• Live Pixel 7 CI/API 36: malformed screenshot payload falls back to a valid 1080×2400 PNG; incomplete input reply retires after one input with dispatched: unknown.
• Our release test app: shell fill/clear and test-IME Unicode fill/clear passed, with backend diagnostics verified.
• Native formatting field: verified adb-shell, correct unconfirmed evidence, exactly one warning.

An earlier local run hit four provider timeouts. The unchanged base, those 14 focused tests, and the final full affected run subsequently passed; the initial cause remains unproven. CI coverage, integration, and native smoke checks remain pending.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.48 MB	5.48 MB	+58 B
Package (unpacked)	5.48 MB	5.48 MB	+58 B
Package (download)	1.66 MB	1.66 MB	+29 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	33.8 ms	33.4 ms	-0.4 ms
CLI --help	66.4 ms	62.1 ms	-4.4 ms

[thymikee]

This PR is ready at 15f17c4. All 16 checks pass with none failing, and the touch-reply and shell-fill paths this diff changes are covered by the CI unit tests. I did not run tests locally. The Pixel 7 and native-field runs in the description are author-reported and I did not reproduce them. Since the diff is a behavior-preserving move, I think they are enough. There are no conflicts, and nothing else stands in the way of merging.

Not blocking: the moved reply parser has three throw branches (wrong protocol, stale requestId, missing blank-line separator), and the protocol test in snapshot-helper-session-protocol.test.ts covers only the stale requestId, so a reorder of the checks would pass the suite; also, the shell loop in text-input.ts still builds the fill evidence and then calls reportAndroidFillUnconfirmed by hand, while other paths go through completeAndroidFillVerification. You can take or leave both. If you take the first, cases for a header-only reply, a wrong protocol with a stale requestId, and a stale requestId with no separator would pin the order.

[thymikee]

Thanks. I am leaving the optional test additions out under the user's explicit no-new-tests constraint. The earlier manual malformed-reply probes cover dispatch disclosure and retirement, but do not pin every parser precedence combination in the permanent suite.

The shell loop keeps evidence construction before completion because it must inspect the unconfirmed result to decide whether its shell-specific retry is allowed. reportAndroidFillUnconfirmed is the common diagnostic/return boundary after that decision; completeAndroidFillVerification uses the same reporter for paths without that retry decision. Routing the shell loop through the latter would either rebuild the evidence or move shell retry policy into shared completion. I am keeping the current separation rather than changing the merged behavior for this optional suggestion.