refactor(commands): read command inputs and output from caller hosts - #3829

Size Report

Startup median (7 runs, lower is better):

Coverage follow-up: 7e2cae20ed fixes the daemon progress fixture by passing explicit files; the failing assertion assumed filesystem directory order. Restacked this unchanged commands patch to fb36192358; range-diff is identical. Foreground check:affected passed at fb36192358 (run-j7njcE, exactCommit=true). CI is running on the pushed head.

Second coverage fix: 288e6cceae polls the detached bridge’s expected cwd contents; the shell creates the evidence file before writing it, and CI read an empty file (run 39289751890). Head 96260a8b76 inherits both this fix and 7e2cae20ed; upper-layer range-diff is unchanged. Foreground check:affected passed (run-bAjCg9, exactCommit=true). Cold self-review found no issue in the test-only diff. CI is running on this head.

The code in 96260a8 is ready to merge. CI is still pending: 17 checks are cancelled or queued, with no failure logs, so there is nothing to attribute yet. The cancelled runs are Smoke, Lint/Typecheck/Package, Repo Guards, Coverage and Integration, and this diff touches the same routes they exercise (src/commands, src/mcp, src/cli output). No conflicts. Before merge, the Coverage, Repo Guards and Smoke lanes need to finish green on 96260a8. Not blocking, and you can take or leave these: captureHostEnvironment(process) in src/io.ts still names the global process in a non-entry module (I could not read the #3823 rule, so please confirm it passes, or take the host as a parameter as #3759 intends); the diff, color and system tests pass process.env and process.cwd() as the host and the MCP and batch paths have no test, so one host-distinct case each would catch a regression to process.cwd(); five CLI sites (progress.ts:133 and :226, reporters/default.ts:146 and :172, proxy.ts:109) still call supportsColor(process.env) instead of using the CLI host, so one output-context owner could decide color once; and formatSnapshotDiffText takes env: EnvMap while formatScreenshotDiffText takes host: HostEnvironment, and PrepareBatchStep in src/commands/batch/projection.ts copies the PrepareDaemonCommandRequest shape and its closure shadows the outer input. I ran no tests, and I did not inspect the MCP executor tests or the reordered host capture in command-tools.ts.

Re-ran Apple 39294051862 once: smoke:form-input failed on fill with typed MAIN_THREAD_TIMEOUT / runner_main_thread_timeout in XCTest; smoke:webview-remote-content then timed out waiting for its link. This layer changes no Swift, so these are suspected runner/fixture flakes pending the one rerun.