> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tashvi.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Skills
> Write a jewelry process down once, like how your studio builds a plain gold bangle, and Create follows it step by step on the designs you pick it for. Make skills by conversation, in Settings, or from a SKILL.md file.
A skill is a process you write down once: how your studio builds a plain gold bangle, your finishing rules, the way you set a halo. Pick it on [Create](https://design.tashvi.ai/create) and Tashvi follows its steps, so you stop retyping the same instructions. A picked skill outranks the preferences Tashvi remembers about you. Only the request you're typing wins over it.
## What a skill holds
| Field | What it is | Limit |
| --------------- | ---------------------------------------------------------------- | ------------------------------------------------------------- |
| **Name** | What you call it, like "Plain gold bangle" | 80 characters |
| **Handle** | What you type in the composer, like `/gold-bangle` | 64 characters: lowercase letters, numbers, and single hyphens |
| **When to use** | One line on where it applies, like "any halo ring, in any metal" | 1,024 characters |
| **Steps** | The process itself | 1,200 characters |
| **Always on** | Applies the skill to every design without picking it | On or off |
| **Enabled** | Off keeps the skill saved but never used | On or off |
Skills are text only, and you can keep up to 50.
## Create a skill
With the skill creator or Settings, nothing is saved until you press **Save skill**. An imported file is saved as soon as it passes the checks.
Type `/skill-creator` in the Create composer, or type `/` and pick **skill-creator** under **Create**. Sent on its own, it asks which pieces the process covers and what it should control. Or describe the process right after the command, like `/skill-creator how we finish every ring`.
Tashvi may ask one more round of up to three questions. Skip any, or close the panel to have it write the skill straight away.
The draft card lists **Things to check**, including any number in the draft you never gave it, like a karat or a width. Ask for changes in words ("make step 3 shorter"), then press **Save skill**. **Use it now** picks it for your next design.
Open your account menu, choose **Settings**, then **Skills**, and press **New skill**. Describe your process in your own words, up to 4,000 characters, and press **Create skill**. Tashvi fills in the name, when-to-use line, and steps.
Every field stays editable. **Handle**, **Always on**, and **Enabled** sit under **More options**. Press **Save skill** when it matches how you work, or choose **Write it yourself** to skip the drafting.
Skills are standard SKILL.md files, the open format other AI tools use for skills. In **Settings → Skills**, choose **Import SKILL.md** and pick one `.md` or plain-text file, up to 32,000 characters.
The file needs `name` (the handle) and `description` (when to use it) in its frontmatter, then the steps. Tashvi's own fields under `metadata` are optional:
```markdown gold-bangle-SKILL.md theme={null}
---
name: gold-bangle
description: Any plain gold bangle, in any karat.
metadata:
tashvi.title: Plain gold bangle
tashvi.alwaysOn: "false"
tashvi.enabled: "true"
---
1. Solid 18k yellow gold, 4 mm wide, with a comfort-fit inner edge.
2. High polish outside, satin finish inside.
3. No stones unless the request names them.
```
An invalid file is refused, with the first problem named. Nothing is fixed silently.
## Use a skill
Type `/` as the first character in the composer. Keep typing to filter your skills, then press Enter or Tab to pick one.
The skill shows as a blue `/handle` token. Write your request as usual, like "a 6 mm bangle with a twisted rope edge."
Under the reply, **Applied:** names the skills that were actually used.
A picked skill stays on for every design you send until you remove it or reload the page. Press Backspace at the start of the composer to remove the last token.
## How skills are applied
* **Up to two per design.** Skills you pick come first. **Always on** skills fill any slot left, most recently used first.
* **Whole or not at all.** Two skills share roughly 1,500 characters of room. If one doesn't fit, it's skipped and left out of **Applied:**.
* **Ahead of inferred preferences.** When you pick a skill, preferences Tashvi inferred on its own sit out that design. Anything you taught Tashvi directly still applies.
* **On new designs, edits, and [plans](/features/create/plan).** A skill you pick while a plan is open shapes the next version of it.
* **Not on a selected part.** Edits made with [Segments](/features/create/segments) or the **Select** tool leave skills out.
* **Not in a private chat**, and never guessed. A skill applies only when you pick it or set it to Always on.
## Manage and share
**Settings → Skills** lists every skill with its badges (**Always on**, **Off**, **Imported**) and how often it's been used. Edit, delete, or download any skill as a SKILL.md file there. Skills belong to your account, so to share one, send the file and have your colleague import it.
Skills are available on every plan, including Free. Making, importing, and downloading them costs no credits, and using one adds nothing to a design's usual [credit cost](/features/create/quality-tiers).
## Troubleshooting
* **The skill is too long to apply.** Shorten the steps or merge short lines. Long lists take more room than their character count suggests.
* **"You already have a skill called /…"** Change the handle under **More options**, or change `name` in an imported file.
* **"You can save up to 50 skills."** Delete one you no longer use.
* **A skill isn't in the `/` list.** Turn **Enabled** back on in Settings.
* **Applied doesn't show my skill.** Two other skills were used, it didn't fit beside another long one, or the turn was a selected-part edit or a private chat.
## Related
See and edit the brief before anything generates. Your skills shape it.
Refine any design with follow-up messages in plain language.