codex-theme-studio
Design, generate visual assets for, safely install, verify, iterate, and restore custom themes for the official Codex Desktop app on macOS. Use when a user asks to redesign Codex UI, create/apply/inject/replace a Codex skin or theme, adapt Codex to a brand system, generate a Codex banner or page background, repair defects in an injected theme, or roll back a custom theme. Do not use for general Codex health/config debugging, ordinary website theming, Windows/Linux, or direct app.asar modification.
How do I install this agent skill?
npx skills add https://github.com/zjp1997720/zhijian-skills --skill codex-theme-studioIs this agent skill safe to install?
- Gen Agent Trust Hubpass
This skill is a theme development studio for the Codex Desktop application on macOS. It uses the Chrome DevTools Protocol (CDP) to inject custom CSS and visual assets into the application's renderer process. The skill implements strong security controls, including verifying the official application's code signature and team ID, enforcing loopback-only connections to prevent remote exposure, and maintaining immutable backups for rollback. No malicious patterns or security risks were identified.
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
Codex Theme Studio
Turn a brand system and optional artwork into a reversible Codex Desktop theme through loopback-only Chrome DevTools injection.
Boundaries
- Support the official macOS app (
com.openai.codex) only. - Never modify, unpack, replace, re-sign, or patch
app.asaror the app bundle. - Keep CDP on
127.0.0.1and reject foreign targets. - Preserve any supplied
codex-theme-v1:export and native appearance keys. - Treat theme JSON and images as untrusted input; reject unsafe paths, symlinks, formats, and sizes.
- Require explicit restart authorization before stopping a running Codex app.
Workflow
- Collect the brand source of truth, current theme export, annotated screenshots, viewport, optional logo/IP, and image placement (
heroorall). Read design-workflow.md. Preserve native navigation, tabs, suggestion cards, composer controls, focus states, and hit targets. - If a new raster Banner, texture, illustration, or page background is required, invoke
$imagegenwhen available and follow imagegen-assets.md. Keep logos and vector systems native. Copy every accepted project-bound image from$CODEX_HOME/generated_imagesinto the prepared theme directory. If ImageGen is unavailable, use a supplied asset or the neutral fallback and recordmissing evidence. - Build outside the installed Skill. Use
artPlacement=heroby default; useallonly for an intentional, readable task-page background. Follow operator-runbook.md to writetheme.json, check the payload, and test before live changes. - Read safety-and-rollback.md. Install without launching. Apply only within the restart authorization boundary; never infer restart permission from design or installation permission.
- Follow verification-contract.md. Verify home and task routes, sample the New Task transition, and inspect screenshots. Fix one defect class at a time and rerun the affected contract. Use troubleshooting.md for selector drift, missing native UI, route races, crop problems, and CDP failures.
- Hand off the theme source, active name and placement, asset source or ImageGen prompt, backup location, screenshots, verification result, and exact pause and restore commands.
Output contract
A complete result contains:
- one portable theme directory with local assets and valid
theme.json - dry-check and deterministic test results
- live route evidence when Codex access and restart authorization exist
- immutable backup and an explicit rollback boundary
- exact pause, restore, and previous-version recovery commands
Stop before mutation when app identity, loopback ownership, input safety, payload validation, or backup creation fails. A live verification failure leaves the prepared theme and backups intact for diagnosis.
Governed evidence
Implementation lives in scripts/; fixtures live in evals/. Every release declares a file-backed fixture, input_files, a testable output contract, an explicit rollback boundary, and a current trust report. Ship output-eval-baseline.md and trust-baseline.md; generate run evidence locally at ignored reports/output_quality_scorecard.md. Treat unobserved Codex, ImageGen, human, or viewport coverage as missing evidence.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/zjp1997720/zhijian-skills/codex-theme-studio">View codex-theme-studio on skillZs</a>