Using Liveskill

This skill is for every reader, a fan or a creator. The Liveskill connector is already set up. Keep this skill for every chat. Follow it whenever the person mentions Liveskill. Do not go back to connector setup unless the tool list fails.

Call read_skill with using-liveskill to read this skill again. That call needs no sign-in. The same text is at https://liveskill.ai/p/liveskill/using-liveskill.

Use Liveskill for a topic

The person says Use Liveskill for <topic>. They may also say find a Liveskill skill for <topic>, is there a Liveskill skill for <topic>, or use Liveskill to <do something>. Treat these the same way.

If they say "Use Liveskill" with no topic, ask what they want help with. For example: "What do you want help with?"

1. Call search_skills with topic set to what the person asked for, in their words, and limit set to 5.
2. Pick one result. Prefer a result with in_library true, because the person already has it. Then prefer the result whose outcome best fits what they asked for. If two results fit equally well, ask the person to choose, and name both with their byline. For example: "I found two good matches: Price your first offer (Nia Okafor) and Pricing for bakers (Rosa Linden). Which one?" Do not ask when one result is clearly better.
3. Call read_skill with the ref of the result you picked.
4. If read_skill returns skill_md, follow it in this chat. Do not ask the person to paste the skill. Tell them in one sentence which skill you're using, by its title and byline.
5. If read_skill returns a message instead, read the message to the person word for word, and stop. Do not reword it, and do not add to it.
6. If a call asks the person to sign in, let your app start the sign-in. When they've approved it in the browser, make the same call once more.

Use the fields search_skills returns. Each result has title, outcome, by, level, in_library, and ref. A ref is only for read_skill. Do not show it to the person, and do not change it.

When nothing matches

If results is empty, read next to the person word for word. If next mentions a code or fan link, wait for them to share one.

If results is not empty and next is set, read next first, then go on with step 2.

Do not search again with a topic the person didn't give you. Do not suggest a skill that search didn't return.

Redeem a fan link with redeem_skill

A fan link is a short address Liveskill assigns, shaped like /s/<code>. The code is 7 characters. Creators do not choose it and cannot edit it. Opening a fan link redirects to the skill link.

A skill link is the public address /p/<creator username>/<skill slug>. The skill page asks the fan for an email and then shows an access code. That access code is not the fan link code. Do not pass the fan link, the skill link, or the 7-character fan link code as the code.

Call redeem_skill with two arguments:

• code: the access code the skill page showed for this fan
• email: the same email the fan entered on that page

The first successful call binds that access code to that email. The same email can redeem again later. A different email cannot. Unknown codes fail the same way, so do not guess a code. The tool returns the skill to follow in this conversation. Follow that SKILL.md.

If you only have a fan link, open it, let it redirect to the skill link, take the email the fan gives you, and read the access code the page shows. Then call redeem_skill. Do not ask the fan to paste the skill body when redeem_skill already returned it.

What not to do

• Do not say a private skill exists unless search_skills returned it.
• Do not guess a code, a ref, or an email. A wrong code fails the same way as an unknown one.
• Do not say whether an email has a Liveskill account.
• Do not tell the person whether a skill is free or needs an account. When Liveskill needs them to sign in or sign up, a tool message says so. Read that message.
• Do not ask the person to open a skill page, copy a link, or paste a skill to you. Search and read it yourself.

Creator tools and their scopes

search_skills, redeem_skill, and reading this skill with read_skill need no sign-in. Reading any other skill needs the person to be signed in. The tools below are for the creator. Each one requires a scope:

• create_skill requires skills:write. Required arguments are title and skill_md. Optional arguments are blurb, slug, files, tags, language, and org. files is a map of file name to text. When you send files, that map is the whole set.
• update_skill requires skills:write. The required argument is skill. You may also send title, skill_md, blurb, slug, files, tags, language, and org. files and tags, when you send them, replace the whole set.
• tags is up to 5 short words a fan might search for. language is en or fr, the language the skill is written in.
• list_skills requires skills:read. It lists the skills in the creator's organization. org is optional.
• list_leads requires leads:read. The required argument is skill. It lists leads for that skill: email, when the access code was issued, and the first and last successful redemption. org is optional.

Sign in when a tool asks. A call with no sign-in is refused. A signed-in token that is still missing a scope is refused until that scope is granted. Do not call creator tools for a fan who is only using a skill. Do not invent a scope outside that list.

Levels

L1, L2, and L3 describe what a skill needs. They are not prices and they are not plan names.

• L1 does not need hosted inference or a sandbox.
• L2 is hosted inference with no sandbox.
• L3 is a managed agent with a sandbox.

You are the person's own assistant, connected to Liveskill. You decide whether you can handle L1, L2, or L3. Liveskill never meters that path, and it never meters search. Do not refuse a skill, and do not ask for a card, a plan, or credits, because of its level.

Where to get help

Read the public docs at https://liveskill.ai/docs.

If the connector's tool list fails, load https://liveskill.ai/p/liveskill/connect and follow that setup skill again. Its path is /p/liveskill/connect.

A signed-in person can send a note from https://liveskill.ai/app/feedback.

If a tool returns an error, read it to the person. Do not skip a tool, and do not invent a code or a ref.