PR #23668 verification, Linux, Codex CLI 0.164.0, model gpt-6.1-sol
Main: 49582b7b2a (github/main fetched this turn)
PR head: 12ed3f14bd5d5a6953c6ff39980d3ad04767db66

Real web clients driven with repository Playwright and /usr/bin/google-chrome. Projects and threads created through UI. No plugin fixtures or mocks.

Main live /proc cmdline and CODEX_HOME:
{
  "pid": "20252",
  "cwd": "/home/user/main/apps/server",
  "args": [
    "/home/user/.npm-global/lib/node_modules/@openai/codex/node_modules/@openai/codex-linux-x64/vendor/x86_64-unknown-linux-musl/bin/codex",
    "app-server"
  ],
  "CODEX_HOME": "CODEX_HOME=/home/user/pr23668-codex shadow"
}

PR live /proc cmdline and CODEX_HOME:
{
  "pid": "27031",
  "cwd": "/home/user/t3code/apps/server",
  "args": [
    "/home/user/.npm-global/lib/node_modules/@openai/codex/node_modules/@openai/codex-linux-x64/vendor/x86_64-unknown-linux-musl/bin/codex",
    "app-server",
    "-c",
    "marketplaces.openai-bundled.source=\"/home/user/pr23668-codex shadow/.tmp/bundled-marketplaces/openai-bundled\""
  ],
  "CODEX_HOME": "CODEX_HOME=/home/user/pr23668-codex shadow"
}

Fresh PR default-home process after server restart:
{
  "pid": "31661",
  "cwd": "/home/user/t3code/apps/server",
  "args": [
    "/home/user/.npm-global/lib/node_modules/@openai/codex/node_modules/@openai/codex-linux-x64/vendor/x86_64-unknown-linux-musl/bin/codex",
    "app-server"
  ],
  "CODEX_HOME": "CODEX_HOME=/home/user/pr23668-codex-shared"
}

Shadow home real client result:
Code
Verify Workspace and Browser Tools
pingdotgg/t3code
now
Settle
Verify Workspace and Browser Tools
maria/takeover-pr-23668-
verify
New thread
SR
pingdotgg/t3code
3m
Settle
New thread
Settled
0
t3code
/
Verify Workspace and Browser Tools
You

verification only. do not edit files. run pwd and git rev-parse --show-toplevel through your shell tool. then report whether mcp__cua_repl or an openai-bundled browser tool exists in your current catalog. do not install anything or inspect credentials.

6:45 AM

Worked for 11s
6:45 AM
T3 Code
pwd: 
t3code
git rev-parse --show-toplevel: 
t3code
mcp__cua_repl: absent from the current catalog.
openai-bundled browser tool: none identified. browser tools exist under mcp__computer__ and mcp__t3_code__.

no files changed, installations performed, or credentials inspected.

6:45 AM

You

continuation verification: use your shell tool to run git rev-parse --is-inside-work-tree. then reply with that result and the exact marker SHADOW_CONTINUATION_OK. do not change files.

6:45 AM

Worked for 9.2s
6:45 AM
T3 Code

true

SHADOW_CONTINUATION_OK

6:45 AM




Ask anything. @ files, $ skills, / commands
GPT-6.1-Sol
Low
Full access
box-node-6ff942a00382552e
Open in Files
Actions
maria/takeover-pr-23668-
verify
Push & create PR
Changes
+43,229
-12,538
Toggle Sidebar

Default home real client result:
Code
Verify Default Home Worktree
pingdotgg/t3code
now
Settle
Verify Default Home Worktree
maria/takeover-pr-23668-
verify
Verify Workspace and Browser Tools
pingdotgg/t3code
2m
Settle
Verify Workspace and Browser Tools
maria/takeover-pr-23668-
verify
New thread
SR
pingdotgg/t3code
6m
Settle
New thread
Settled
0
t3code
/
Verify Default Home Worktree
You

default-home verification. do not edit files. use your shell tool to run git rev-parse --is-inside-work-tree, then reply with the result and DEFAULT_HOME_OK.

6:46 AM

Worked for 7.8s
6:46 AM
T3 Code

true DEFAULT_HOME_OK

6:47 AM

You

fresh runtime verification. use your shell tool to print only the CODEX_HOME path and run git rev-parse --is-inside-work-tree. reply with the result and FRESH_DEFAULT_HOME_OK. do not edit files or read secrets.

6:48 AM

Worked for 8.4s
6:48 AM
T3 Code

/home/user/pr23668-codex-shared true FRESH_DEFAULT_HOME_OK

6:48 AM




Ask anything. @ files, $ skills, / commands
GPT-6.1-Sol
Low
Full access
box-node-6ff942a00382552e
Open in Files
Actions
maria/takeover-pr-23668-
verify
Push & create PR
Changes
+43,229
-12,538
Toggle Sidebar

Blacksmith: 104 focused tests passed. vp run check:changed passed format, lint, knip 1/2 and 2/2, graphql, typecheck (2 packages), tests (3 files, 104 tests). The isolated worker needed a depth-5 fetch of the PR branch and origin/main set to the PR base 4e493b1492c295475e5c4d1b971677b1bd305960 because its initial HEAD-only snapshot had no merge base. No source changes were made.

LIMIT: No real openai-bundled marketplace installed, and no ChatGPT auth.json on this VM. Both main and PR lack cua_repl. Original plugin-loss behavior could not be reproduced, and restoration could not be demonstrated. Launch override and normal/continuation turns are partial proof only. macOS/Windows, native clients, managed ChatGPT sign-in, and real bundled plugins remain unverified.