Codex App cannot execute Docker container tests despite Docker being available on the host

Open 💬 2 comments Opened Aug 8, 2026 by Edydh

What version of the Codex App are you using (From “About Codex” dialog)?

Powered by Codex & OWL Version 26.803.41515

What subscription do you have?

Pro

What platform is your computer?

Darwin 27.0.0 arm64 arm

What issue are you seeing?

Codex App is unable to complete Docker-based test execution even though Docker is running and available on the host Mac.

The failure occurs when Codex tries to run or validate tests that depend on Docker containers. Codex can inspect the project and reason about the required test, but the Docker execution step does not complete successfully.

The behavior is repeatable in the same environment. We are still collecting the exact command output, Codex App version, Docker version, and host platform details so we can add a precise reproduction sequence and error text.

What steps can reproduce the bug?

  1. Open the Codex App on macOS with a local project that uses Docker-based tests.
  2. Start Docker Desktop and verify that Docker is running normally on the host.
  3. Open the project in Codex App.
  4. Ask Codex to run the project’s Docker-based test/validation workflow.
  5. Codex inspects the project and identifies the required Docker test operation.
  6. When Codex attempts to execute the Docker-dependent test, the operation does not complete successfully.
  7. Repeat the same request in the same environment; the Docker execution failure is reproducible.

What is the expected behavior?

Codex App should be able to execute the Docker-based test workflow when Docker is running and accessible on the host Mac.

If the required Docker command is valid and the containers are available, Codex should run the test, capture the result, and report whether it passed or failed.

If Codex cannot access Docker because of sandboxing, permissions, or another environment restriction, it should clearly explain the limitation and the exact reason instead of failing to complete the Docker step.

Additional information

Docker Desktop is running normally on the same Mac, and the affected workflow is a legitimate local development/test workflow using project-owned containers.

The issue appears specific to Codex App’s ability to execute or complete the Docker-dependent step, rather than Docker being unavailable generally.

We searched the existing Codex GitHub issues using terms including docker container, docker exec, and docker compose and did not find a clear duplicate matching this behavior.

We can provide additional diagnostics, including the exact failing command/output, Codex App version, Docker version, docker compose version, and uname -mprs output, after reproducing the issue again on the affected Mac.

No customer production environment is required to reproduce the problem.

View original on GitHub ↗

2 Comments

Edydh · 20 days ago

Docker environment:

  • Docker Desktop: 4.85.0 (235549)
  • Docker Engine: 29.6.2
  • Buildx: v0.35.0-desktop.2
  • Docker Compose: v5.3.1
  • Credential Helper: v0.9.8
  • Kubernetes: v1.36.1
Edydh · 19 days ago

Update after further reproduction

I’ve gathered additional evidence that changes the initial diagnosis.

Docker Desktop itself is working correctly and Codex App can successfully use Docker in this environment.

Since opening this issue, Codex has successfully:

  • created isolated PostgreSQL 15.8 and PostgREST containers;
  • executed SQL/runtime tests;
  • run multi-session concurrency tests;
  • run production-shaped performance benchmarks;
  • remove the disposable containers/networks afterward;
  • preserve an unrelated local Docker/Supabase stack untouched.

So this does not currently appear to be a general inability for Codex App to access Docker.

The reproducible problem is closer to task/session instability during longer-running Docker/database review workflows.

Observed behavior has included:

  • Task encountered a system error after extended analysis;
  • failures after automatic context compaction/resume;
  • a resumed shell temporarily losing the Docker CLI from PATH even though Docker Desktop was running;
  • the task stopping before completing a multi-session PostgreSQL/PostgREST concurrency experiment;
  • similar system errors occurring even during a later static/read-only review where Docker was explicitly not used.

After updating Docker Desktop and restarting with smaller scoped tasks, Codex again successfully executed Docker-based PostgreSQL validation and performance testing.

Current Docker environment:

  • Docker Desktop 4.85.0 (235549)
  • Engine 29.6.2
  • Buildx v0.35.0-desktop.2
  • Compose v5.3.1
  • Credential Helper v0.9.8
  • Kubernetes v1.36.1

This suggests the useful reproduction target may be long-running Codex task reliability / context-compaction recovery, rather than Docker access itself.

I’m happy to provide additional logs if there is a preferred diagnostic path.