# Takeover verification — PR 8668

Original: [#8668](https://git.cafe/pingdotgg/t3code/pulls/8668), @limwa's `feature/fix/linux-protocol-launcher`, original head [fa3d27b](https://github.com/limwa/t3code/commit/fa3d27bf1c1fd2a7d42f3afacbb98efebedef488).
Replacement: `takeover/8668-linux-launcher`, integrated head [f42e44a](https://github.com/pingdotgg/t3code/commit/f42e44a1e5b3508add3ab8818e0b8e3956e39a57), baseline [3cc31ed](https://github.com/pingdotgg/t3code/commit/3cc31ed27c8956b36cee35a50509fb7c6869592b).

## Tests

Two original PR regression tests were rewritten to use a launcher path containing spaces. No production test behavior was removed. Current main's 17 tests in the two writer suites become 19 (DesktopLinuxUrlHandler:13→14; DesktopPreReadyPlatform:4→5); the original PR also had 19 in those files. Both new tests fail against main because APPIMAGE wins, and pass with the fix. Opus reviewer independently repeated that failure check.

Final scoped run includes DesktopEnvironment.test.ts: 21/21 tests across three files. Desktop typecheck, six-file lint, formatting, and git diff --check pass. A full desktop build (web, server, desktop assets/bundles) also passes. No repository-wide tests/typecheck were run.

## Independent review round 1

GPT-6.1 Sol and Claude Opus5.5 each reviewed the full six-file diff at c5e3300148 against main909e4c9482 and returned MERGE with no code findings. Both identified runtime evidence as the remaining gate. Main was then merged by commit f42e44a; all six reviewed files remain byte-identical, confirmed with git diff --exit-code. No source implementation changes followed reviews.

Both writers use launcher→APPIMAGE→execPath, preserve existing escaping and platform/registration gates, and keep blank/unset behavior. No new regex, parser, input loop, dependency patch, wire contract, renderer change, or sensitive surface. Existing string trim is linear in the environment-value length, called once during setup.

## Process

No fork push, rebase, force-push, or PR merge was performed. The replacement has focused tests and implementation commits on current main plus an ordinary main merge. No fixture/evidence/scratch files enter the PR. All runtime state is fresh private /tmp state; no live T3 database was read or copied.

Models/harness: Claude Opus5.5 via T3-owned Claude agent for tests and implementation; GPT-6.1 Sol via Codex and Claude Opus5.5 for independent reviews; GPT-6.1 Sol via Codex for proof audit and runtime evidence; lead GPT-6.1 Sol via Codex in T3 Code.