# Runtime evidence for [#33105](https://git.cafe/pingdotgg/t3code/pull/33105)

Before: [4c361199](https://git.cafe/pingdotgg/t3code/commit/4c361199f47f6ea32a2c14835df392640d0f7c63); after: [1bc13bdd](https://git.cafe/pingdotgg/t3code/commit/1bc13bddf86515d310344be3e7e4820a851fa1bb). Linux 6.8.0-117-generic x86_64; Node v26.12.1; Claude Code 2.1.321. Same installed executable on both sides. Real T3 dev servers used fresh revision-owned databases under this scratch directory, started with `T3CODE_DEV_SEED=0`. No live database or credentials were copied.

Only `CLAUDE_CODE_ENTRYPOINT`, `CLAUDE_CODE_ARTIFACT`, and `claude_code_artifact` are recorded from child environments; JSON null means absent. Wrapper PID/parent and CLI child PID are synthetic-run process identifiers. All model identifiers, session/account identifiers, unrelated tools and output are removed. Original stdout remains private. These are actual process and init captures, not test expectations.

Claude was reached through its native CLI with a synthetic API credential and a deliberately unreachable loopback API endpoint. No model response or successful authentication is claimed. The wrapper forwards CLI stdout unmodified and terminates its owned CLI after observing init; code 143 is that intentional stop, not auth failure. C2 capability probes have control initialization but no prompt/init tool list. C3 uses JSON output, with a five-second owned-child stop after spawn-env capture; no answer is required. Artifact counts are exactly the names `Artifact`, `ArtifactComments`, `ArtifactData` in real `system/init` events.

AI-assisted; verified with the tests and tools listed above.


# Windows limitation and existing focused tests

W1, W2 and W3: **cannot-verify windows**. This VM is Linux; no native Windows spawn was attempted. The existing W1 unit assertion runs the helper with HostProcess.Platform=win32. W2 shares that branch. W3 is covered at the generic instance-merge unit layer by the auditor’s source review; it was not rerun here. This test result is supplementary branch evidence, not a replacement for a native runtime capture.

Executed at head: `vp test run apps/server/src/provider/Drivers/ClaudeHome.test.ts`.

```text

note: You are running `vp test` as a Vite+ built-in command. If you meant to run the test npm script, use `vpr test` instead.

 RUN  v5.0.5 /home/user/wt/pr-33105


 Test Files  1 passed (1)
      Tests  5 passed (5)
   Start at  07:24:47
   Duration  10.77s (transform 69%, setup 26%, import 5%)



```