Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

chore(store-submissions): regenerate the OpenAI store submission draft

nirholas3ee286803542
diff --git a/prompts/store-submissions/_generated/openai-submission.md b/prompts/store-submissions/_generated/openai-submission.md
index 0b6043aa9..478a5729f 100644
--- a/prompts/store-submissions/_generated/openai-submission.md
+++ b/prompts/store-submissions/_generated/openai-submission.md
@@ -791,17 +791,27 @@ that promises a fixed turnaround.
### Step 6. The demo recording
-The one asset that does not exist yet. The form wants a video of the plugin working, recorded
-in ChatGPT Developer Mode. Screen recordings are acceptable and OpenAI states it is not shared
-externally; it is used to validate the test cases above.
-
-1. Enable Developer Mode in ChatGPT.
-2. Add `https://three.ws/api/mcp-studio` as a custom connector. No credentials.
-3. Record positive case 1 end to end, with the inline viewer and the Download control both
- visible in frame.
-4. If the generation runs long, cut the waiting rather than speeding up the footage. A sped-up
- recording implies a turnaround a reviewer will not get, which is the one way this video can
- work against the submission.
+The form wants a video of the plugin working, recorded in ChatGPT Developer Mode. Screen
+recordings are acceptable and OpenAI states it is not shared externally; it is used to validate
+the test cases, so it must show every one of them behaving as the Testing tab says.
+
+1. Enable Developer Mode in ChatGPT (Settings, Apps, Advanced).
+2. Add **`https://three.ws/api/mcp-chatgpt`** as a custom connector, no authentication. Not
+ `/api/mcp-studio`: that surface waits up to three minutes inline, which ChatGPT cuts off at
+ 60 seconds, and it carries the persona tools the listing leaves out.
+3. Record, in one conversation, in this order, with the prompts pasted exactly as in the Testing
+ tab:
+ 1. Case 1 (robot mascot). Show the viewer's timer running, then the model landing on its own.
+ 2. Case 3 (make the shell matte), straight after case 1. Show the Versions strip, and click
+ Original and back.
+ 3. Case 4 (the armchair URL). Show the frames and the armchair in the viewer.
+4. In a second conversation, record case 2 (the knight): the viewer finishes the mesh, rigs it by
+ itself, and the chip reads Rigged Model.
+5. In a third, record case 5 (the telescope through the art-directed generator), then the three
+ negative prompts, showing that none of them calls the plugin.
+6. Cut the waiting rather than speeding it up. A sped-up recording implies a turnaround a
+ reviewer will not get. Leave the viewer's elapsed timer visible on each side of the cut so the
+ real duration stays honest.
### Step 7. Global