@adeildo/pi-skills

What changes

This piece brings no skills of its own. It changes how you call the ones pi already loaded, from your skill folders or from packages.

Pi expands /skill:name only at the start of a message, and only one. With this piece, any /skill:name in the text counts:

use /skill:simplify, then go over the docs with /skill:unslop

Before the message, the model gets one block per skill, the same <skill> block pi writes for /skill:name. Your text reaches it as you typed it. In the chat, each reference shows as [skill] name on a violet background, the way pi labels a folded skill. The blocks stay folded until Ctrl+O, which expands tool output, shows them in full.

A reference counts when it names a skill pi loaded and follows a space, the start of a line or an opening bracket. Inside backticks it stays text.

Typing a reference

A / past the start of the message opens a list of the skills alone, narrowed as you type, the way pi's palette does at the start. Picking one writes /skill:name and a space. The list opens on / in the editor of @adeildo/pi-look, which also colors each reference while you type. In any other editor, Tab after the / opens it.

At the start of the message, / still opens pi's own palette, with the commands and the skills.

Notes

• A message that starts with /skill: goes out with a space before it, so pi does not expand that skill a second time. The chat does not draw the space.
• While the model works, the skills queue with the message, as a steer or a follow-up.
• If a skill file cannot be read, the message goes without that skill and a warning says why.

Part of pi-harness, which brings every piece in one install.