agentleFS
appautomatonappautomaton/delineate/.agents/skills/climate-garden/SKILL.md

climate-garden

Skill3 starsChanged 30 days agoMITView on GitHub

Make or change a garden on this project's Garden site (garden/): the plants of one real climate, drawn in code as watercolor-and-ink SVG that grows as the page scrolls, by day and by night. Use for a new garden or climate, a plant added, redrawn, or refined, a garden's composition changed (sides look empty, plants crowd, the sky fills up), seasons or water given to a garden, or a plant's look researched, even if the user never says "skill" or names the files.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Real plants, drawn true: Every plant exists and is grown in the climate, wild or cultivated.

    line 22
  2. Very specific:

    Every task ends the same way: npm run check and npm run look <key> (both draw the plants first), then read the screenshots and tell the …

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Climate garden

Project paths are relative to the repository root; ref/ files are in this skill's folder. Load a ref when its step comes up, not before.

The bar

A garden a visitor can tell from every other by silhouette alone, before color: about twelve real plants of one climate, drawn as one hand would draw them, growing in as the page scrolls, by day and by moonlight.

  • Real plants, drawn true. Every plant exists and is grown in the climate, wild or cultivated. How it looks is researched before it is drawn; where it is native is not our question.
  • One hand. A new plant is drawn in the manner of the plants around it, at the same weight of wash and ink, so the garden reads as one sheet.
  • It grows. The page is a growth, not a picture: limbs extend, then parts unfold where the limb has reached. The first scroll must visibly answer.
  • Nothing floats. Whatever hangs (lichen, a vine, spent flowers, fruit) hangs from a host drawn with it. Side plants reach in from beyond the frame; bed plants root below it, so no ground is ever drawn. Open land leaves the top to the sky.
  • Judged at garden size. A detail that reads in a close study can vanish where the plant is placed; a flower under about eight pixels is lost.

How a garden works

  • A plant is a function. It draws in its own units, base (or hanging point) at the origin, from its own seed, into a list of SVG elements wrapped in growth markup. The page drives one progress value from 0 to 1; a garden that shows a year maps it onto the calendar.
  • Color is tokens. A drawing names a species tone or a material, never a color; each has a day value and a night value in garden/src/tokens.css.
  • Layout is CSS. Each garden has a stylesheet that places its plants for each tier of screen; a solver then slides them apart.
  • npm run draw renders every plant and fails when a part would open before its limb reaches it. The page, the gallery, and the solver all read what it writes.

To see what exists, read the code rather than a list: the gardens in garden/src/garden/gardens.ts (the page, the switcher, and the gallery find a garden through it), the plants in garden/src/garden/plants.json (names, the gallery's words, rank, and paint order), and the drawings in garden/draw/, where the registry in draw.ts maps each plant to its function. Most gardens draw in a module named for them; the oldest draw in the shared kit. garden/climates.md holds the mechanisms gardens share (Seasons, Water) and each garden's botanical reference.

The task

  • A new garden. Research the roster (ref/botany.md) and decide it: about four sides, two or three hangers on their hosts where the climate has a canopy (none on open land), five or so in the bed. Decide the moment or the year: a climate whose plants change their silhouette through the year (wood that goes bare, herbs that die back) earns a year; one whose silhouettes hold (evergreens, succulents, cushions) shows its best moment. Record it in climates.md. Wire the garden in wherever an existing one is wired (see Finding the places, below), its gardens.ts entry first. Start with the signature plant alone, draw it until it sets the manner (ref/drawing.md), and show it to the user before the rest. Then draw the rest in one batch, judge each, compose (ref/layout.md), and verify (ref/verify.md).
  • A plant added to a garden. Research the one plant. Check its silhouette against the garden's others before drawing. Read its neighbors' code and draw in their manner, with a close study beside the drawing for the gallery. Wire it in wherever a neighbor is wired: its registry entry (seed, drawing, close study) and its plants.json entry together (npm run draw stops on a plant listed but not registered), its token in both lights, its place in the layout tiers where it earns one, and its rows in climates.md.
  • A plant redrawn or refined. Find it through plants.json and its garden's module; read its entry in climates.md. Judge it in the gallery by day and by night and at garden size before and after the change.
  • A garden recomposed. ref/layout.md, then npm run look for every tier.
  • Seasons or water. The mechanisms are in climates.md; timing parts on the year is in ref/drawing.md.

Finding the places. Lists of files go stale; the code does not. git grep -il <key> for an existing garden's or plant's key, together with the files named for it, shows every place one is wired in; a new one goes in each of those places. Prose that counts the gardens (the README, delineate's page) names no key: search for the count in words.

Every task ends the same way: npm run check and npm run look <key> (both draw the plants first), then read the screenshots and tell the user where to look, what changed, and what is still off.

Working rules

  • The roster, the drawings, and the layout are yours to decide. Decide, build, show; the user corrects. Ask only where the choice is theirs (which garden next, a plant they may not want).
  • Draw back to front, and build volume from parts: a mass painted behind a shrub reads as a rock.
  • Reuse the kit before writing geometry. When a second garden needs a helper that lives in one garden's module, move it into a shared module then, named by what it is, not by the plant it first served.
  • Improve rather than preserve: an earlier drawing is not a specification. When an approach fails, delete it; keep only code the drawings use.
  • A lesson another garden will need goes, as a rule, into the ref it belongs to (a drawing trap into ref/drawing.md, a placement into ref/layout.md). What only one garden needs lives in its code.

Routing

StepLoad
Researching a roster or a plant, briefing research agentsref/botany.md
Drawing a plant, growth markup, tokens, the emblem, a drawing that looks wrongref/drawing.md
Placing plants, tiers of screen, sides, crowdingref/layout.md
Running the site, looking at layouts, the buildref/verify.md

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.