[Fix] Shell exit codes are always missing from task transcripts and the completion check

[mrubens]

Problem

OpenCode's shell tool (1.18, which the worker pins) reports a command's exit code as metadata.exit. extractOpenCodeExitCode in the harness reads metadata.exitCode, metadata.code, and the same two keys on the output object, so every shell result is recorded with a null exit code. On the nightly deployment, 0 of 3,283 shell results in the last 24 hours carry one. The shell tool's own <shell_metadata> block only reports timeouts and aborts, so the code is not recoverable from the output text either.

This matters more since #3071: the completion check holds the agent's "tests pass" claim against the recorded commands, and without exit codes it can only go by the output tail.

Change

Read metadata.exit as well, ahead of the legacy keys. One line plus a test that feeds the 1.18 metadata shape through the harness and checks the exit code reaches the completion check.

Validation

• Harness completion-gate tests (13), pnpm lint:fast, pnpm check-types:fast.

[roomote-community]

No code issues found. See task

_{Reviewed 9b6275e}