Start with tt-a1i/archify if you ever explain architecture changes to a team. You describe the system, it produces interactive HTML and SVG diagrams, and it can diff two snapshots so you show exactly which components and connections changed between versions. A sensible first experiment: describe the current design of a small service, then describe the proposed redesign, and use the comparison view in your next review meeting instead of a slide full of red boxes you drew by hand.
For something lighter, bilawalsidhu/gods-eye-view puts flights, ships and earthquakes on one 3D globe in your browser. Click a plane and follow it across the terrain. Keep in mind that the night vision and thermal styles are visualization effects, not real sensor imagery, so treat it as a fun way to explore geographic data rather than a monitoring tool.
If your struggle is prompting images rather than building systems, freestylefly/awesome-gpt-image-2 is a collection of image generation examples for posters, product shots and interface mockups. Instead of staring at a blank prompt box, pick a layout you like and swap in your own subject and content. The first experiment writes itself: recreate your product poster in three different styles from the collection and keep the one that reads at thumbnail size.
Now the two discipline tools. DietrichGebert/ponytail is a set of instructions you give a coding agent: reuse existing code first, question whether a new abstraction is needed, and never drop validation or error handling. mattpocock/skills focuses on understanding the request, with small editable workflows where the agent interrogates you about the requirement before writing any code. Ponytail restrains the agent’s output, mattpocock/skills fixes the input. Pick ponytail if your agent keeps inventing frameworks for a two-line change; pick the questioning skill if it starts coding before you finished explaining.
If you want to run several agents at once and see the mess, stablyai/orca is a desktop app that runs terminal coding agents in separate Git worktrees so you can compare attempts at the same task side by side. Try this: give the same bug to three agents in orca, review each diff, and notice which one you would actually merge. Keep the prompts and compare where the implementations differ; those differences give you something specific to ask about in the next attempt.