Como usar o Claude como planejador de jardim: o guia do Skill Ruben

Yes, for the parts that are structuring, sequencing and remembering: turning your own observations into a bed layout, planning succession so a bed keeps producing, scaling ambition to the space and time you actually have, and keeping a record from one season to the next. Not for the facts about your specific plot, which it does not have. The method is to supply what it cannot know and forbid it from filling the gaps.

No, and confusing them is the most common error in AI gardening advice. The USDA Plant Hardiness Zone Map, updated in 2023, is built from the average annual extreme minimum winter temperature, shown as 10-degree Fahrenheit zones and 5-degree half zones. That tells you whether a perennial survives your winter and nothing about when your spring starts or how long your season is. Two places in the same zone can have last frosts six weeks apart. Sowing calendars run off frost dates, which are a separate dataset.

No, and this is the important one. It does not have your frost dates, but asked for them it will produce a plausible-looking answer anyway, and your whole calendar then hangs off a guess. Look up your last spring frost and first autumn frost once, from your national met service or a local extension or gardening body, write them down, and hand them to the model as facts it must use rather than something it should estimate.

Because a calendar written in month names carries a silent assumption. Ask without saying where you are and you get a Northern Hemisphere, usually US, calendar, and it will not tell you it assumed that. For a garden in Australia, New Zealand, South Africa or southern South America every month is roughly six months out while the advice reads perfectly. Also, outside the US the zone system is different: the UK uses RHS hardiness ratings. Insist on weeks relative to your frost date rather than month names.

Treat it as a list of things to look up rather than rules. Some companion planting has evidence behind it and a great deal of it is garden folklore repeated as fact. A model has read both and cannot tell you which is which, so it produces a confident pairing table with no distinction between the tested and the traditional. A useful instruction is to make it label every pairing as either well supported or traditional, and to say when it is not sure.

Check the packet. These vary by variety, the packet in your hand is the authority, and invented cultivar names are a common failure. A better pattern is to have the model write a placeholder such as CHECK YOUR PACKET or SPACING: packet, then recalculate the bed layout once you give it the real numbers from the seed you actually bought.

No. Half a dozen different problems look identical in a written description, and it cannot see the plant, the soil or the weather it has had. A local nursery, a plant clinic or a gardening society will do better in two minutes. This goes double for anything you intend to eat, for foraging identification, and for any decision about applying a chemical, all of which need a qualified local source.

The download is a ZIP with SKILL.md at the root of the archive, not inside a nested folder. In the Claude desktop app open Customize then Skills, upload the ZIP and toggle it on. Skills need code execution enabled, under Settings then Capabilities. Anthropic's help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise, so if you are on the free plan check Settings then Capabilities for your own account rather than trusting either page. In ChatGPT or Gemini, paste the contents of SKILL.md into custom instructions instead.