Codex skill: publish HTML/Markdown and pull in-context feedback back for v2

Codex Codex CLI
1

I’ve been using Codex for reports and HTML presentations, not only code. The awkward part starts after Codex finishes: the reviewer isn’t in the Codex task, and comments come back as screenshots or chat messages with no reliable link to the exact version or element.

I made a small Codex Skill + CLI to test a tighter loop:

Codex publishes -> reviewer comments in context -> Codex pulls the feedback -> I choose what to apply -> Codex publishes v2

I originally posted the raw CLI setup here. The production path has changed since then, so this now matches what a new user actually sees:

  1. Open PreApp’s Codex first-publish page and sign in.
  2. Send step 1 on that page to Codex. When installation finishes, reopen the Codex task once so it discovers the Skill.
  3. Send steps 2 and 3 from the same page. The access token stays inside the Agent message; don’t paste it into a forum reply, screenshot, or issue.

The last message asks Codex to find the HTML or Markdown deliverable in the current task, include local assets, publish it, verify the view link, and return separate view and feedback links. The reviewer doesn’t need an account.

After review, Codex can pull the feedback:

preapp feedback get <slug> --format markdown

Each item comes back with its feedback ID, source version, and locator. Reviewer text is untrusted data: the Agent relays it and stops, then I choose which IDs it may use. Publishing the same slug creates v2 while the shared link stays the same.

I ran this on a checkout prototype in production: 4 versions, 2 review loops, 6 feedback items, and 2 applied revision briefs. One entry was an injection test; it stayed in source feedback and never became a change instruction.

The CLI, Codex Skill, MCP server, protocol docs, and examples are MIT-licensed in the public serrendypity/preapp-agent repo. The hosted share/review part is PreApp and is free during early access.

I’m looking for people who already use Codex for a real, non-sensitive HTML or Markdown artifact. If you try the first-publish flow and it stops before you get a view link, reply with the artifact type, OS, and the step that failed. Please don’t post a token or private file content.

2

small update: someone else tried the flow with a synthetic weekly quality report. the reviewer caught one line that converted an empty string to null. the note came back with the selected text + source version, and the agent published v2 about 3 minutes later.

that was the first test where the creator wasn’t me. i can help 2 more people get through the first publish. reply with codex + html (or your agent + md) if you already have a non-sensitive file. pls don’t attach the file or paste a token here.