Awesome Codex Spark

Codex Spark

[](https://github.com/KingGyuSuh/awesome-codex-spark/actions/workflows/test.yml) [](LICENSE) [](https://nodejs.org) [](https://developers.openai.com/codex)

Codex plugin for delegating concrete Computer Use and Browser Use tasks to GPT-5.3 Codex Spark subagents with structured handoffs and auditable traces.

The plugin packages one main-session skill, **`$codex-spark-delegate`**. The parent session keeps scope, approval, model effort, and recovery; a spark subagent runs one bounded UI/browser task as the executor and returns an auditable trace.

**Quick install:**

codex plugin marketplace add KingGyuSuh/awesome-codex-spark

Then `/plugins` → install `codex-spark` → restart Codex. See [Install](#install) for tag-pinned, local-clone, and Codex desktop app paths.

**Contents** — [Why](#why) · [What This Is Not](#what-this-is-not) · [What It Installs](#what-it-installs) · [Requirements](#requirements) · [Install](#install) · [Usage](#usage) · [Validation](#validation) · [Troubleshooting](#troubleshooting) · [Development](#development) · [Documentation](#documentation)

Why

Codex Spark is useful when the main session should not spend its attention on mechanical UI work: opening a page, reading visible state, filling approved form values, clicking through a bounded flow, or checking a local browser surface.

The child agent must return a trace: what it saw, what it did, which checks passed, what artifacts exist, and what the parent should do next. That trace is load-bearing because the parent has to resolve login problems, approval gaps, mismatches, or partial side effects.

What This Is Not

Intentionally narrow: