Skip to content
Skills Library
Design & UX · Planning

codex-visual-handoff

/codex-visual-handoff

Division of labor for users who run both Claude and OpenAI Codex. Use whenever a task produces user-facing visuals — images, video frames, slide decks, rendered pages, thumbnails, covers, motion graphics. Claude owns facts, copy, approval flow, and substance QA; Codex owns everything visual, including writing the image/video prompts. Use this skill when the user asks to "make a reel/thumbnail/cover/deck", when a pipeline mixes text and rendered assets, or when deciding which model should handle a generation step.

Licence
MIT
Last repository change
18.08.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 18.08.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Division of labor for users who run both Claude and OpenAI Codex. Use whenever a task produces user-facing visuals — images, video frames, slide decks, rendered pages, thumbnails, covers, motion graphics. Claude owns facts, copy, approval flow, and substance QA; Codex owns everything visual, including writing the image/video prompts. Use this skill when the user asks to "make a reel/thumbnail/cover/deck", when a pipeline mixes text and rendered assets, or when deciding which model should handle a generation step.

Triggers from the description

make a reel/thumbnail/cover/deck
Original author description

Division of labor for users who run both Claude and OpenAI Codex. Use whenever a task produces user-facing visuals — images, video frames, slide decks, rendered pages, thumbnails, covers, motion graphics. Claude owns facts, copy, approval flow, and substance QA; Codex owns everything visual, including writing the image/video prompts. Use this skill when the user asks to "make a reel/thumbnail/cover/deck", when a pipeline mixes text and rendered assets, or when deciding which model should handle a generation step.

How it works

  1. Why the prompt line matters
  2. The split
  3. The handoff brief
  4. Rejection routing
Original SKILL.md headings · 52 lines, 396 words
SKILL.md preview
---
name: codex-visual-handoff
description: Division of labor for users who run both Claude and OpenAI Codex. Use whenever a task produces user-facing visuals — images, video frames, slide decks, rendered pages, thumbnails, covers, motion graphics. Claude owns facts, copy, approval flow, and substance QA; Codex owns everything visual, including writing the image/video prompts. Use this skill when the user asks to "make a reel/thumbnail/cover/deck", when a pipeline mixes text and rendered assets, or when deciding which model should handle a generation step.
---

# Codex Visual Handoff

One rule, strictly kept: **every rendered visual is Codex's job — including the prompt.**

## Why the prompt line matters

The tempting failure mode is "Claude writes a detailed image prompt, Codex just runs it."
Don't. A prompt is where visual judgment happens — composition, restraint, what to leave
out. If the orchestrator writes it, the output's visual ceiling is the orchestrator's, and
the bill arrives later: the render fails human taste review, gets rejected, and the whole
piece is remade. In a print-pipeline audit of 52 revision rounds, taste rejections were the
single largest rework cause (13 rounds, 9 of them repeats) — and every one arrived *after*
all mechanical checks had passed.

Hand the judgment to the model that's better at it, whole.

## The split

| Step | Owner | Notes |
|---|---|---|
| Facts, numbers, sources | **Claude** | Open the real data/code; never invent |
| Exact copy (user-facing text) | **Claude** | Read it aloud; would a person say this? |
| Image/video **prompt writing** | **Codex** | Claude never drafts it "to help" |
| Asset choice, composition, layout | **Codex** | |
| Rendering, motion, sound mux | **Codex** | |
| Approval flow, lineage hashes | **Claude** | Approved copy hash must match render input |
| Substance QA (facts, copy, gating) | **Claude** | Aesthetic QA goes to an independent reviewer |

## The handoff brief

Claude passes a **requirements brief**, never a design:

- The approved exact copy (verbatim, hashed)
- Facts and figures with sources
- Hard prohibitions (rights-unsafe assets, banned claims, brand invariants)
- Required disclosures
- The role of each deliverable (hook frame / evidence frame / CTA frame)

What the brief must NOT contain: image prompts, palette picks, composition sketches,
"maybe put X on the right". If you catch yourself describing pixels, you're doing
Codex's job with Claude's eyes.

## Rejection routing

- Copy is wrong or awkward → back to Claude.
- Visuals are rejected (alignment, clutter, taste) → **back to Codex.** Claude does not
  hot-patch a rendered asset; a local fix by the wrong model becomes the next rejection.
- Same cause rejected twice → stop adjusting at this level; go up one (copy or structure),
  or ask the user one interpretation question. Repeated-cause rounds were over half of all
  rework in the audit above.

Versions

The database has no other skills named codex-visual-handoff.

Database snapshot: 13.11.2026 · package version 571f761fd166 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 82codex-handoff
    交接当前工作到新 Codex 任务。
  • 100codex-imagegen
    Use this skill when an agent must generate or edit raster images through Codex, especially from Claude Code or another harness without native image generation. Activates on mentions of generate image with Codex, delegate image generation, Codex imagegen, delegated image editing,…
  • 100image-generation
    Optimizes image generation prompts using Subject-Context-Style structure. Use this skill when generating images, creating illustrations, photos, visual assets, editing images, or crafting prompts for any image generation model.
  • 100codex-image-studio
    Image generation workflow for Codex using OpenAI-compatible APIs. Use when the user wants Codex to create, edit, or batch-produce images through a third-party or self-hosted image endpoint, including text-to-image, image-to-image, multi-reference edits, mask edits, or multi-job generation.
  • 100design
    House design standard, the process, anti-slop rules, exact values and local checkers every visual job follows. Use for writing, restyling or reviewing anything visual (websites, landing pages, apps, dashboards, settings, components, CSS, games, celebrations, 3D, slide decks,…
  • 100lookbk
    Find and install a Lookbk visual skill. Use when the user wants to make a launch video, product film, kinetic title, website hero, landing page, slide deck, OG or link-preview image, or diagram "with a Lookbk skill", asks to "find a Lookbk skill", or asks which skill can render…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.