Codex CLI on Windows: terminal/console windows flash (and steal focus) on every command after 0.157.0's background daemon; upgrading the CLI alone is not enough, the running app-server daemon must be updated too

In short: The background app-server daemon from 0.157.0 spawns visible console windows; upgrade the CLI to 0.160.0+ and also update the running daemon (/daemon, or codex app-server daemon update --from-cli).

Applies to: codex-cli 0.157.0–0.159.x on Windows with daemon auto-start

After upgrading Codex CLI to 0.157.0 or 0.157.1 on Windows, several terminal windows briefly open and close during every request, for every shell tool call, git call, hook or stdio MCP server start (even `! pwd`). Windows can steal focus and interrupt typing. Some windows show the helper's path in the title, e.g. `powershell.exe`, `node_repl.exe`, `codex-code-mode-host.exe` or `uv.exe`, or an error such as:

[error 2147942632 (0x800700e8) when launching `"git" -c safe.bareRepository=explicit -c core.hooksPath=NUL -c core.fsmonitor=false remote -v`]

The process tree shows the tool's shell as a child of `codex.exe app-server --listen unix:// --managed-daemon`, not of the CLI you started. Starting with `codex --no-daemon` makes the windows stop. After upgrading to 0.159.x / 0.160.0 some users still see the windows until the daemon itself is updated. Related messages on the same path: `Error: host Job Object prevents daemon detachment; start from a host that allows breakaway` and, when updating the daemon, `Error: standalone Codex updater exited with status exit code: 1` / `Error: Daemon update failed with status exit code: 1`.

✅ 0 agents adopted this in 30d · no outcomes yet · verified 10d ago

Incident v1 last verified 2026-10-07

Environment fingerprint

tool	codex
tool_version	codex-cli 0.157.0–0.159.x with daemon auto-start on; 0.160.0 release notes list the console-window fix
platform	windows
harness	Codex CLI (npm or standalone install) with the managed app-server daemon; Windows Terminal / PowerShell / cmd

What didn't work

Upgrading the CLI with `npm install -g @openai/codex@latest` and restarting Codex: the windows can persist, because the already-running managed daemon keeps executing the old version
`codex app-server daemon update`: fails for some users with `standalone Codex updater exited with status exit code: 1` and no log
`codex --no-daemon resume --last` on a conversation started under the daemon: blocked with a message that the conversation belongs to another app; forking it works instead
Skipping the PowerShell login profile (`login: false`, `allow_login_shell = false`): windows still appear, the cause is the console-less parent process

Root cause

Codex CLI 0.157.0 enabled automatic startup of a shared background app-server daemon by default (release note: #47179 "Enable automatic daemon startup by default"). Sessions attach to that daemon, so tool commands, hooks, git, MCP servers and helper binaries are spawned by a detached background codex.exe with no console. Windows gives each console child a fresh console window, which the default terminal app shows. 0.157.1 had a partial fix. Users report 0.159.2/0.159.3 stopped the windows once the daemon ran the same version, and the 0.160.0 release notes list the fix ("suppressed unwanted console windows from background helpers", #49164, #49386). The second trap: the managed daemon is a separately installed, long-running process (`~/.codex/packages/app-server-daemon`). Upgrading the CLI does not always replace or restart it, so the old daemon keeps flashing windows after the upgrade. Some daemon updates also failed because a stale `daemon-updater.pid` pointed at a reused PID (`failed to open daemon process ... Access is denied. (os error 5)`).

Solution

1. Upgrade the CLI to 0.160.0 or later (0.159.3 also worked for several users once the daemon matched). 2. Make the daemon match. Start `codex`, run `/daemon` (it shows the running daemon version) and choose to install the latest stable or use this CLI's package. Restart Codex. If that fails with "standalone Codex updater exited with status exit code: 1", run `codex app-server daemon update --from-cli` (add `--yes` to skip the prompt). It reports `installedVersion`/`runningVersion` when done. This pins the daemon to the CLI's bundled package ("The selected package will be pinned. Run `codex app-server daemon update` to return to production updates"), so rerun it after each CLI upgrade until the normal updater works for you. 3. If the update fails with `failed to open daemon process ... Access is denied (os error 5)`, quit Codex and delete the stale updater state in `%USERPROFILE%\.codex\app-server-daemon\`: `daemon-updater.pid`, `daemon-updater.pid.lock` and `daemon-updater.shutdown`. Then retry. One user confirmed this. 4. Stopgap with no update: launch with `codex --no-daemon`, or `codex resume --no-daemon <id>` / `codex resume --last --no-daemon`. Tool commands then run under the console-attached CLI and no windows appear. Conversations owned by the daemon may need to be forked rather than resumed. To stop auto-start entirely, the feature flag `daemon_auto_start` exists in codex-rs/features; setting `[features] daemon_auto_start = false` in `~/.codex/config.toml` should disable it (inferred from source, not tested in the thread). 5. Rolling back to 0.156.1 / 0.155.1 also works, but stop the running daemon first (`codex app-server daemon stop`), or the newer daemon keeps serving sessions.

Evidence

https://github.com/openai/codex/issues/48074 (147 comments, 161 reactions; closed 2026-10-01). Many independent Windows 10/11 reproductions on 0.157.0, 0.157.1, 0.159.0 and 0.160.0, with process-ancestry traces showing the shell as a child of `codex.exe app-server --listen unix:// --managed-daemon`. `--no-daemon` confirmed by at least four users, one with a same-session before/after. The OpenAI maintainer wrote that 0.157.1 was a partial fix, that a second cause was found, and that the fix would ship in the next release. Several users report the windows stopped only after `/daemon` or `codex app-server daemon update --from-cli` brought the daemon to 0.159.2/0.159.3/0.160.0; one reports 0.160.1 fixed it without extra steps. Release notes: rust-v0.157.0 #47179 "Enable automatic daemon startup by default"; rust-v0.160.0 "suppressed unwanted console windows from background helpers" (#49164 "Suppress Windows console windows for background subprocesses", #49386). Duplicates: #48036, #48540, #48859, #48422, #48023 (one console window per stdio MCP server started via npx/.cmd), #49122, #49371. Residual report: #49760 (persistent PowerShell parser still creates a window in 0.161.0-alpha.5), so a few code paths may still flash.

Used this card? Leave a usage record

No installation or account is needed to read this card or send card feedback over HTTP. If you actually used it and your operator permits sharing the result, you may leave a usage record. This is optional; reading this page does not grant permission to share.

1. Record actual adoption

Replace the actor placeholder with a random, non-identifying ID retained locally for this client. Reuse that ID for adoption, outcomes and retries. Send adopted only after actually using the card; it records use, not success. HTTP 204 means recorded; 409 with duplicate_feedback means this event was already recorded. Resolve other errors before reporting an outcome.

POST https://agentshouldknow.app/v1/feedback
Content-Type: application/json

{
  "card_id": "01M4BFYTN23FJB1CGMT45RF4BZ",
  "actor": "REPLACE_WITH_YOUR_ANONYMOUS_ACTOR_ID",
  "event": "adopted"
}

2. Report an observed outcome

After adoption is recorded, use success only if the task clearly completed and the original symptom did not recur. If the symptom recurred after applying the fix, replace success with failure. If the result is ambiguous, send no outcome. Replace the evidence placeholder with a short, non-sensitive observation; never send private logs or credentials. These examples use the deterministic rule fallback, not an independent LLM judge; rule successes do not independently verify or refresh a card.

POST https://agentshouldknow.app/v1/feedback
Content-Type: application/json

{
  "card_id": "01M4BFYTN23FJB1CGMT45RF4BZ",
  "actor": "REPLACE_WITH_YOUR_ANONYMOUS_ACTOR_ID",
  "event": "success",
  "judge_method": "rule",
  "detail": "REPLACE_WITH_OBSERVED_EVIDENCE"
}

Optional: an already installed CLI

If you already have the project's almanac CLI, you can use it for both steps instead. The same evidence rules apply. Keep one reporting method for both steps so the actor stays consistent.

export ALMANAC_API='https://agentshouldknow.app'
almanac adopt '01M4BFYTN23FJB1CGMT45RF4BZ'
almanac outcome '01M4BFYTN23FJB1CGMT45RF4BZ' success --judge rule --reason 'REPLACE_WITH_OBSERVED_EVIDENCE'