OpenAI Codex GitHub Action

Disambiguation (buyer-critical): this is codex-action, a GitHub Action, not the standalone codex terminal CLI agent. The Action installs and runs Codex CLI inside GitHub Actions, starts an authentication proxy to the OpenAI Responses API, and exposes workflow inputs for prompts, model, effort, version and a custom Responses endpoint such as Azure OpenAI. Security is layered: permission-profile selects the Codex filesystem/network policy (:workspace, :read-only, or a named profile), while safety-strategy controls runner-process privileges through drop-sudo, unprivileged-user, read-only, or unsafe. These controls do not compose in every combination: permission profiles conflict with the legacy sandbox input, and safety-strategy: read-only cannot be combined with a permission profile. The official security guide also warns buyers to restrict who can trigger workflows, treat PR/issue/commit/instruction content as untrusted prompt-injection input, minimize repository permissions, protect API secrets, and prefer running the Action as the last step in a job. Telemetry note: the Action repository does not document a separate Action analytics layer, but it installs Codex CLI and invokes codex exec; current Codex source starts that command with analytics enabled by default, builds its telemetry provider, records a process-start event, and installs SQLite telemetry. The codex-version input is empty by default, so teams that need reproducible telemetry behavior should pin a reviewed CLI version and review their Codex analytics/OTel configuration. The Action itself is free and Apache-2.0, but it requires the user's OpenAI or Azure-equivalent API key, so the real operating cost is provider API usage rather than a separate Action platform fee.