Exploring Codex v0.128.0: One‑Click /pet Companion and /goal Task Automation

Codex has been updated to version 0.128.0, introducing two notable features: /pet, a one‑click visual companion, and /goal, a command for setting and managing task goals.

/pet – a floating status indicator

The /pet command is available in the Codex desktop app (not the CLI). Typing /pet in the input box spawns a small animated animal in the corner of the screen. The same effect can be achieved via Settings → Appearance → Pets , where a few built‑in pets are offered.

Users can create custom pets with the /hatch skill. The steps are: $skill-installer hatch-pet After installing, press Cmd + K to open the command palette, run Force Reload Skills , and then generate a pet, for example:

$hatch-pet create a new pet inspired by my recent projects

Images in the original article show a custom pet generated from recent projects and the floating pet in action.

According to the official documentation, the floating window remains visible even when switching to other applications, displaying the currently active thread and a short progress hint. Thus, /pet functions as a work‑status indicator, turning the anxiety of checking a background agent into a quick glance at a cute animal.

/goal – persistent task‑loop automation

The /goal command is a new CLI feature that must be enabled in the configuration file:

[features]
goals = true

or via the CLI:

codex update
codex features enable goals

OpenAI’s leadership described it as a built‑in “Ralph loop++”. The original Ralph loop concept repeatedly runs the same prompt in a while loop until the task is completed, previously requiring third‑party plugins such as ralph‑loop or ralph‑wiggum . The /goal command incorporates this pattern natively and adds several upgrades.

Key improvements include persistence: goals are stored in a thread_goals table, so progress survives client disconnects or thread restarts. This makes long‑running tasks feasible without keeping a terminal window open.

The TUI supports four actions: /goal <objective> – create a goal /goal pause – pause execution /goal unpause – resume execution /goal clear – clear the goal and start over

These controls let users pause an agent that is veering off‑track, adjust the prompt, and then continue, avoiding the need to abort the process manually.

Model tools have also been added: get_goal (query current goal), create_goal (create a new goal), and update_goal (mark completion). Permissions are deliberately limited so that only the user or runtime can pause, resume, or clear a goal.

User comments highlighted that the momentum often drops after each agent turn; /goal addresses this by outsourcing the iterative prompting to the agent itself, reducing the need for manual context switching.

The author concludes that the update follows a “serious‑on‑the‑left, playful‑on‑the‑right” strategy: /goal provides a solid engineering upgrade by embedding the proven Ralph loop pattern, while /pet offers a low‑cost, attention‑grabbing way to surface status information.

Developers are advised to prioritize /goal for immediate productivity gains, using /pet only if the visual novelty is desired. The next preview version, 0.129.0‑alpha, is already listed on the release page, indicating a rapid release cadence.