oneshotlm
Compare all models on this prompt →

Maze generator & solver · gpt-5.6-sol

Generate a random maze with the recursive-backtracker algorithm on a canvas, then animate an A-star or BFS solver finding the path from start to finish.

⛨ sandboxed Open in new tab ↗
transcript · 6 messages
user

Generate a random maze with the recursive-backtracker algorithm on a canvas, then animate an A-star or BFS solver finding the path from start to finish.

assistant
bash {"command":"ls -la && find . -maxdepth 2 -type f -print"}
toolResult

total 0 drwxr-sr-x. 2 node node 6 Aug 3 00:22 . drwxrwsrwx. 3 root node 16 Aug 3 00:22 ..

assistant
thinking
**Planning maze generation and animation**

I’m working on a maze with cell walls and a backtracking approach. I’ll implement the A* algorithm for the cell graph, allowing for visual elements where the frontier exploration is animated in colors like cyan and blue, while visited paths are purple and final paths gold. I also want a clear UI legend and metrics. It’s important to manage canvas dimensions accurately and ensure a smooth animation experience. Let's see how I can streamline the generation and solving processes!**Optimizing A* pathfinding and animation**

I’m considering how to manage the A* process while recording visits and snapshots. I want to visually represent the visited and open sets during each step and compute these as expanded nodes with newly discovered ones. I’m planning to use the Manhattan heuristic and have BFS as an option. I also need to think about features like pause and resume for the animation, as well as adjusting canvas sizing based on activity and maintaining the correct aspect ratio. Let's nail down those dimensions!**Creating UI Layouts**

I’m thinking about different compact sizes: 21x14, 31x20, 41x27, and 51x34. I want them to fit nicely in the UI with the top mast, control toolbar, and a status chip on the canvas. I should consider side stats and a guide too. It's important to avoid oversized text, so I’ll keep that in mind as I create the layouts. Overall, I want everything to flow smoothly and look neat!
write index.html (17289 bytes)
toolResult

Successfully wrote 17289 bytes to index.html

assistant

TAKE IT TO CODE

A single self-contained HTML file — copy it and open in any browser, or reuse the prompt with your own pi-agent run.

RUN

modelgpt-5.6-sol
vendorOpenAI
harnesspi-agent
generated2026-08-03

EVALUATION

3/5 partial

Maze generator/A* solver visualizer; carving and pause states show but no active pathfinding progress observed, minor UI glitches on scroll zoom.

Vision grade · anthropic/claude-sonnet-5

MEASUREMENTS

Raw readings from the capture, not a verdict — the grade above is the only judgement. Percentages are how much of the screen changed; the measure under-reports sparse motion on a dark background.

idle motion0.1%
on click3.1%
on drag2.7%
on wheel2.5%
on enter + space0.4%
on arrow keys34.2%
on w/a/s/d0.0%
frame spread18.5 / 255
console errors0
js errors none

METRICS

runtime81.1s
tokens in2.4k
tokens out6.3k
cached2.3k
cost / run$0.1897