Codex

Connect Codex to YouThumb to prepare thumbnails from your agent workspace: upload logos and screenshots, write the prompt, launch the generation and collect the results. The skill supplies the API contract and the rules that keep your credits under control.

Install the skill for Codex

Run the installer from the workspace you use with Codex:
bash
npx skills add https://github.com/mikecodeur/skills --skill youthumb-api
Choose Codex when prompted. To select it explicitly:
bash
npx skills add https://github.com/mikecodeur/skills --skill youthumb-api --agent codex
Open a new Codex session in that workspace and ask it to use the youthumb-api skill. The skill describes the integration; your runtime still needs network access and a YouThumb API key.

Connect your YouThumb account

In YouThumb, open Account → API Keys and create a key. Make it available as YOUTHUMB_API_KEY in Codex's runtime:
bash
export YOUTHUMB_API_KEY="your-api-key"
The key is for YouThumb, not an OpenAI API key. Keep it out of committed files. A hosted agent environment needs its own secret configuration; it does not inherit your local shell variables.

Prepare your assets

text
Use the youthumb-api skill. Check that ./assets/logo.png and
./assets/screenshot.png are real PNG images, upload them as assets
and give me their URLs. Do not create a project yet.
The skill asks the agent to verify each file before uploading: an HTML page saved as .png would fail.

Create the project, then generate

text
Create a thumbnail project with the free preset: me on the left,
@image1 (logo) on the right with a soft glow, @image2 (screenshot)
in the background. Title "SHIP IT". 2 variations.
Show me the detailed status before starting.
Codex checks that both references are attached, then starts the generation when you confirm. Each variation consumes credits, so start small.

Collect the results

Ask Codex to poll the status until it is completed, then download the images into your project folder. Polling too fast triggers the rate limit: one request at a time, with a pause between calls.

Troubleshooting

  • Skill not found: check the installation scope and start a new session where the skill was installed.
  • 401: check the key and whether it was revoked. The API is strictly rate limited, so a 401 right after many calls can also mean the agent should wait before trying again.
  • 402: your credits are insufficient for the generation.
  • 429: too many calls; ask the agent to wait 30 to 45 seconds instead of retrying in a loop.
See Agents and the Developer API.