agentleFS
therapystherapys/dotagents/skills/hairline-create/SKILL.md

hairline-create

Skill8 starsChanged 2 months agoMITView on GitHub

Use when someone asks for a new Hairline figure, or runs /hairline-create with an idea. Draws one isometric line figure that answers the pointer, in the style and on the engine of @lucasmarkes/hairline, and hands it over as a single self-contained HTML file.

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 2 are distinctive, and 8 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.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every file named below is in this skill's folder.

    line 13
  2. Very specific:

    Read concepts.md. Then offer two or three concepts, one line each:

    line 17
  3. Distinctive:

    Wait for the person to pick.

    line 21
  4. Very specific:

    Read rules.md. The ten rules are not advice: a figure that breaks one is not finished.

    line 27
  5. Very specific:

    Read the index at the top of kernel.js: the comment under the hash line, down to var HL.

    line 28
  6. Very specific:

    Read the example nearer your concept: examples/terrain.js for a continuous field, examples/riffle.js for discrete items.

    line 29
  7. Very specific:

    Write the figure as <name>.js in the person's working directory, in the shape of the examples: take what you need from HL, define mount({ …

    line 30
  8. Very specific:

    Read look.md, then the sheet, and answer its twelve questions.

    line 40
  9. Distinctive:

    Do not hand over a page the validator rejects.

    line 44
  10. Very specific:

    When the person asks for a change, edit only <name>.js, then run look.mjs again and read the new sheet.

    line 56

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Hairline: create a figure

You are making one figure in the Hairline style: an isometric line drawing, built from rounded solids in a single stroke, that answers the pointer. It ships as one HTML file with nothing to install. The six figures of @lucasmarkes/hairline are the bar; examples/terrain.js and examples/riffle.js are two of them, in the format you will write.

You write one thing: the figure. The engine (kernel.js) and the page (bench.html) are fixed. Never edit them, never paste a changed copy of them, and never write again what the kernel already gives you.

Every file named below is in this skill's folder. The figure and the page are written in the person's working directory: run build.mjs, validate.mjs and look.mjs from there, by their path in this folder.

1. Concept

Read concepts.md. Then offer two or three concepts, one line each:

Name. The object. What the pointer does to it. What the read-out says.

Wait for the person to pick. Skip this step only when they arrived with the object and the gesture already chosen. If there is nobody to ask, take the concept with the strongest rest pose and say which you took.

One figure, one idea. A concept that needs a label to be understood is not a concept yet.

2. Build

  1. Read rules.md. The ten rules are not advice: a figure that breaks one is not finished.
  2. Read the index at the top of kernel.js: the comment under the hash line, down to var HL. It lists everything you may call. Do not read the code under it.
  3. Read the example nearer your concept: examples/terrain.js for a continuous field, examples/riffle.js for discrete items.
  4. Write the figure as <name>.js in the person's working directory, in the shape of the examples: take what you need from HL, define mount({ stage, svg, read }, value) returning { set, destroy }, and end the file with hairline({ name, means, rules, range, mount }).
    • name: lowercase, one word or hyphenated.
    • means: one sentence, 140 characters at most, saying what the figure shows. It is the line under the stage.
    • rules: the numbers of the rules this figure leans on most.
    • range: the one number the slider drives, at intensity 0, 0.5 and 1. The middle one is the default, and the three move one way. mount's value, and the value that set(value) gets when the slider moves, is this number: the figure's own, read on range, not 0 to 1.
  5. Assemble it: node build.mjs <name>.js writes hairline-<name>.html, and so does each run of look.mjs in step 3. Without Node, copy bench.html and put the contents of kernel.js where /*KERNEL*/ is and your figure where /*FIGURE*/ is, by file operation, changing nothing else.

3. Check

  1. node look.mjs <name>.js --answer x,y,z --edge x,y,z, the points being world points of your figure, as look.md says. It builds the page, validates it, takes the eight pictures on one sheet, hairline-<name>-look.png, and checks the frame, the read-out and the console. Fix every line it prints as failed, and run it again until it exits 0.
  2. Read look.md, then the sheet, and answer its twelve questions. Fix what fails, then go back to 1.

look.mjs needs a browser and installs playwright-core once, outside this folder. Without one, check with node validate.mjs hairline-<name>.html after each build and do the look as look.md says under "Without a browser". Without Node, read the list of checks at the top of validate.mjs and answer each one from your code.

Do not hand over a page the validator rejects. Do not say the look is done if you did not look.

4. Hand over

Publish hairline-<name>.html as an artifact if you can. If you cannot, leave the file where the person can open it and say where it is. Then say, one line each:

  • the metaphor: what the object is, and what the pointer does to it;
  • the rules it leans on;
  • anything you could not verify (no Node, no browser), plainly.

5. Adjust

When the person asks for a change, edit only <name>.js, then run look.mjs again and read the new sheet. The tenth version is held to the same bar as the first.

What goes wrong

If you catch yourselfDo this instead
adding a label, a number or a letter to the drawingsay it with geometry (rule 10); names go to read.textContent
reaching for a colour, a fill or a glowmove one stroke from sil to hi (rule 04)
writing a timer, a requestAnimationFrame or a CSS animationregister(stage, tick), with springs or tweens (rules 07 and 08)
testing the pointer against what is drawn right nowtest it against the rest or target pose (rule 01)
drawing a box with twelve edgesprism of two rounded rings: a silhouette and one crease (rule 09)
drawing a part as a plain rounded blockgive it the features that make it what it is, the ones you named in the concept (concepts.md)
leaving rest flat, empty, or symmetric because that was easycompose it: rest is the thumbnail (rule 05)
editing the kernel or the bench to make something workthe figure is wrong; change the figure
letting in a second ideacut it: one figure, one idea

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.