flow-graphics
Skill1 starsChanged 3 months agoMITView on GitHub
Draw, animate, and record graphics from a Flow program using the native gfx backend. Use for windows, games, visualisations, animation, GIF recording, shaders, or any request to see a simulation rather than read its numbers.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- flow-audioSkill
- flow-autodiffSkill
- flow-basicsSkill
- flow-compiler-devSkill
- flow-debugSkill
- flow-dynamicsSkill
- flow-effectsSkill
- flow-perfSkill
- flow-setupSkill
- flow-stdlibSkill
- flow-verifySkill
- flow-wasmSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 103
When the user asks to see something, record it and give them the file:
- Distinctive: line 132
Do not claim a visual result you have not rendered.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Graphics
Flow draws through a native backend: Cocoa on macOS, SDL2 on Linux and Windows, canvas in the browser. The same source runs live, records deterministically, and compiles to WebAssembly.
Requires FLOW_HOST=python.
The three ways to run it
FLOW_HOST=python flow gfx program.flow # live window
FLOW_HOST=python flow window program.flow # SDL2 window
FLOW_HOST=python flow record program.flow --frames 200 --gif out.gifrecord runs headless and writes PPM frames, or a GIF with --gif. It takes
--skip, --out DIR, --fps, --stride, --width, and --keys SPEC for
scripted input. Use it whenever you need to show a result rather than watch it,
and in any environment without a display.
The API
import "stdlib/gfx.flow"
import "stdlib/keys.flow" # named key codes
function main() -> i32 {
let g: Gfx = gfx_open(760, 460, "Title")
while !gfx_should_close(g) {
gfx_poll(g)
if gfx_key_down(g, KEY_ESC) { break }
gfx_clear(g, 18, 22, 18)
gfx_fill_rect(g, 100, 100, 60, 40, 30, 160, 60)
gfx_present(g)
gfx_wait_frame(g, 60)
}
gfx_close(g)
return 0
}| Function | Does |
gfx_open(w, h, title) -> Gfx | Open the window |
gfx_poll(g) | Pump events |
gfx_should_close(g) -> bool | Window closed |
gfx_key_down(g, keycode) -> bool | Key state |
gfx_clear(g, r, g2, b) | Fill the frame |
gfx_fill_rect(g, x, y, w, h, r, g2, b) | Solid rectangle |
gfx_blit_rgb(g, x, y, w, h, src) | Blit a raw RGB buffer |
gfx_present(g) | Show the frame |
gfx_frame_pump(g) -> bool | Poll, close check, and Esc check in one call |
gfx_wait_frame(g, fps) | Pace the loop |
gfx_time_ms(g) -> f64 | Milliseconds since the window opened |
gfx_close(g) | Shut down |
Colours are i32 in the range 0 to 255. Coordinates put the origin at the top
left.
Time behaves differently per backend, on purpose
gfx_time_ms is monotonic real time in a window, virtual in the recorder
(exactly one frame per present), and performance.now() in the browser. A
recorded run is therefore identical on every machine. gfx_wait_frame paces a
live window and does nothing in the recorder.
Pace with gfx_wait_frame rather than a sleep. Without it, a demo runs as fast
as the machine allows and its speed becomes hardware dependent.
Text
stdlib/text.flow is the most imported module in the repository. Use it rather
than drawing glyphs from rectangles.
grep -n "^export function" lib/stdlib/text.flow | head -20Read before writing
ls examples/games/ # snake, tetris, 2048, breakout, invaders, pong
ls examples/morphogenesis/ # reaction diffusion, growth
ls examples/graphics/ # gif writer, shaders, text
ls examples/planet/ examples/procgen/examples/games/snake_gfx.flow is a complete, readable program with a title
screen, input edge detection, and a game over state. Copy its structure.
Recording is the deliverable
When the user asks to see something, record it and give them the file:
FLOW_HOST=python flow record examples/games/snake_gfx.flow \
--frames 200 --gif snake.gifFrames go under $FLOW_HOME/build/frames_<name>/. The GIF goes where --gif
says. Report both paths and the frame count that was actually written, which
the recorder prints.
Shaders and 3D
FLOW_HOST=python flow shader program.flow # Metal fill shader, macOS
FLOW_HOST=python flow demo vulkan basicstdlib/render3d.flow, stdlib/vulkan_renderer.flow, and
docs/language/shaders.md cover the rest. Vulkan on macOS runs through
MoltenVK.
Pitfalls
- A window needs a display. In a headless session, use
record. - Key codes are platform values. Import
stdlib/keys.flowrather than typing numbers, and say which platform a raw code came from. - Sample input once per frame and compare against the previous frame for edge detection. Reading a key twice in one frame gives inconsistent results.
- Do not claim a visual result you have not rendered. Record it and look at the frame count.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

