Codex App: Revamped Automations with Execution Environment Control

Automations Rebuilt: Isolation, Model Control, and Templates

With version 26.312, OpenAI redesigned the Automations feature inside the Codex app, transforming it from a basic scheduling tool into a configurable background-task system suitable for team use. The changes address three distinct developer needs: execution safety, model tuning, and workflow discovery.

Choosing Where Automations Run

The most consequential addition is the new execution environment selector. Each automation can now be configured to run in one of two modes:

Local execution runs the automation directly in the project's main checkout, with full read/write access to the working tree. This is appropriate for lightweight read-only tasks (generating a summary, answering a question) or for projects not tracked by Git.

Worktree execution — available for Git repositories — runs the automation in a dedicated background worktree that is fully isolated from the developer's current changes. Each scheduled run starts fresh in its own branch. There is no risk that a nightly automation creates merge conflicts, touches uncommitted work, or leaves stray files in the main checkout. Codex manages worktree cleanup automatically (configurable on or off in Settings), so unused worktrees do not accumulate over time.

This is how professional CI systems have always operated — isolating automated work from the developer's active context — and bringing the same model to scheduled Codex tasks makes unattended automation genuinely safe.

Per-Automation Model and Reasoning Level

Previously, automations used whatever default model and reasoning settings were active globally. The revamped interface exposes these as per-automation settings that can be left on defaults or configured explicitly:

• Model selection: choose any model available in the account (e.g., GPT-5.4 for intensive analysis, a faster lower-cost model for routine checks)
• Reasoning effort: set to low, medium, or high, trading compute cost for depth of analysis

This granularity matters for cost-conscious engineering teams. A daily "summarize open issues" automation does not need the same reasoning depth as a weekly "security review of all new commits" — and now each can be tuned appropriately.

Automation Templates

A new templates library accompanies the redesigned interface. Rather than beginning with a blank prompt, developers can browse pre-built starting points for common recurring workflows:

• Daily issue triage (label, prioritize, and summarize open issues)
• CI failure monitoring (summarize recent failed builds and surface fix suggestions)
• Release brief generation (compile merged PRs since the last tag into a readable summary)
• Commit-level bug checks (scan new commits for common error patterns)

Templates are fully editable before scheduling — they function as informed starting points rather than locked-in configurations.

Skills Integration

Automations can be paired with Skills using the -name syntax, enabling complex multi-step workflows to be packaged and reused across different automation schedules. For example, a skill that queries a project management tool can be wired into a daily triage automation without duplicating the underlying logic.

The Bigger Picture

The revamped Automations feature positions the Codex app as a viable background-task orchestrator for small to mid-size software teams. By combining worktree isolation with per-task model control and a templates library, OpenAI offers a lightweight alternative to custom CI jobs that requires no infrastructure setup and runs wherever the Codex app is installed. The result is a system where recurring, important-but-tedious developer work — issue triage, alert monitoring, CI/CD summaries — can be delegated with confidence.