Grok Bots generate game art asset sheets and transparent PNGs

• A specialist bot called Picasso turns moodboards into style-matched game asset sheets, checks visual consistency, then crops the results into same-size transparent PNGs, according to DannyLimanseta's Picasso walkthrough.
• The workflow is already feeding a playable MMO-dev-studio prototype, while DannyLimanseta's graphics note identifies Grok Bot and Grok Imagine as the source of its graphics.
• A separate bot called Mirai summarizes Slack, packages reported bugs, and sends a Cursor-generated pull request back for approval in DannyLimanseta's Mirai walkthrough.
• The setup is a small agent roster rather than one overloaded chat: DannyLimanseta's bot roster places a Chief of Staff above specialist bots and skills.
• User-submitted bug reports remain the awkward boundary. levelsio's bug-board concern describes keeping a feedback board disconnected from an agent over prompt-injection worries.

xAI's skills documentation defines a skill as reusable instructions that capture steps, decision rules, expected output, and safety boundaries. Picasso's visible output includes a zip of eight separate quest-giver assets, while Mirai's interface shows an issue becoming a reviewable Cursor PR.

DannyLimanseta's system uses a Chief of Staff bot to spin up specialists and attach task-specific skills.

That maps cleanly onto the product's design: xAI's launch post says Bots work on their own computers inside existing apps, while its skills guide says a saved skill can be shared across Bots when the required connector or login is present. A broader minchoi roundup collected examples of the same multi-bot pattern across operations, research, and production work.

Grok Bot remains an early beta. xAI's August 21 availability update lists SuperGrok Plus, Cursor Pro+, and Cursor Teams plans.

Picasso begins with reference gathering, then produces a sheet designed around one chosen art style.

The pipeline has five explicit stages:

1. Find moodboards for the target style.
2. Generate assets together in a sheet.
3. Use Grok Imagine 2.0 for the images.
4. Run a skill that evaluates every asset against the style.
5. Crop the approved assets into transparent PNGs at consistent sizes.

The output format is the interesting bit. It moves from image generation to a game-ready archive, including an eight-character quest-giver zip and a later revision that names a ranger's body type in the prompt. DannyLimanseta's confirmation calls it the art pipeline currently in use.

The image model is swappable within the same process. For Cook the Dungeon, DannyLimanseta's Artgen note says the bot uses a custom Artgen tool rather than Grok Imagine.

Mirai is connected to a work Slack channel and produces a daily summary of notable conversations.

When a teammate reports a bug, the sequence is:

1. Summarize the report.
2. Start a Cursor agent on the fix.
3. Return a pull request for DannyLimanseta's approval.

The attached task card records a completed image-edit cropping-logo fix and exposes the PR in Cursor. DannyLimanseta reported that this loop had reduced some fixes to under 10 minutes.

The asset workflow has a concrete destination: a weekend prototype where players develop a fledgling MMO through a tech tree, buildings, NPCs, and mobs.

Game UI remains a manual tuning problem. In DannyLimanseta's UI note, he said models were still weak at game UI and that getting it right took substantial pixel-pushing through prompts.

The Mirai flow ends with a human approval step. In a separate coding workflow, levelsio's terminal workflow shows bug reports being copied manually into an agent running on a VPS rather than flowing in from a connected board.

Levelsio's concern is a feature request that instructs an agent to add a backdoor while disguising the change as a routine pagination fix. marckohlbrugge's reply proposed reviewing the original report before an agent begins work, and levelsio's response called that a possible fix.