Feature request: First-class workflow artifacts for reusable Codex work

Problem

Codex can help design and execute useful workflows today, but many of the valuable outputs stay trapped in a conversation. When Codex produces a plan, script, checklist, prompt, test harness, generated document, verification routine, or repeatable agentic flow, there is not a clear first-class way to save it as a durable, reviewable, reusable artifact.

This makes it harder to:

• Re-run a useful workflow later
• Review generated scripts or task plans before execution
• Share a successful flow with a teammate
• Attach an artifact to a repo, task, issue, or PR
• Track which scripts or flows were generated for a specific objective
• Turn a successful Codex conversation into a repeatable project asset

There is already a related request for user-defined Codex workflows in openai/codex#24721. This request is related, but slightly different: workflows describe a repeatable procedure; workflow artifacts are the durable outputs and bundles Codex creates, edits, saves, reruns, and updates as part of that procedure.

Proposed Solution

Add support for first-class workflow artifacts in Codex.

An artifact would be a structured, inspectable output created or maintained by Codex, such as:

• A runnable script
• A workflow plan
• A generated prompt or skill draft
• A test or verification harness
• A checklist
• A generated issue or PR template
• A small utility or task runner
• A multi-step agentic flow definition

Codex should let users create, review, edit, save, update, and re-run these artifacts from the Codex UI or workspace.

Example Workflows

Generate a reusable script

The user asks Codex to investigate a repo health check. Codex creates a script artifact with commands, expected outputs, and safety notes. The user reviews it, saves it to the workspace, and can re-run or revise it later.

Turn a conversation into a workflow bundle

The user works with Codex to triage GitHub issues. Codex converts the successful process into a reusable artifact bundle with inputs, steps, commands/tools used, output files, and review checkpoints.

Create a project-specific verification artifact

Codex fixes a bug and creates a verification checklist or test harness. The artifact can be attached to the PR or reused for similar changes.

Acceptance Criteria

• Codex can create a named artifact from a conversation.
• Artifacts are inspectable before execution or saving.
• Artifacts can include structured metadata such as title, type, inputs, steps, commands, safety notes, and expected outputs.
• Users can edit artifacts before saving or running them.
• Artifacts can be saved into the workspace or associated with a repo/thread.
• Runnable artifacts clearly show what commands or tools will be used before execution.
• Codex can update an existing artifact instead of regenerating it from scratch.
• Artifacts can be referenced later in the same project or thread.

Safety Considerations

• Runnable artifacts should require clear user review before execution.
• Codex should distinguish between drafts, saved artifacts, and executed artifacts.
• Artifacts should not silently expand permissions or bypass sandbox constraints.
• Generated scripts should include visible assumptions and expected side effects.
• Sensitive data should not be embedded into artifacts by default.

Non-goals

• This is not a request for Codex to automatically execute arbitrary generated scripts.
• This is not a request to replace existing repo files without review.
• This is not a request for a general-purpose low-code app builder.
• This is not a request for hidden background automation without user approval.

Why This Matters

A lot of valuable Codex work is not just the final code change. It is the reusable process Codex discovers along the way: scripts, checks, task flows, prompts, and verification routines. Making these first-class artifacts would help users turn one-off agent conversations into durable project assets.