Skills Search: Real Life Scenarios

127 documents. Commands, skills, custom notes. Here’s how I actually find what I need.

Scenario 1: “Prime Number” — Found by Keyword

I had a theoretical note about prime number architecture. I don’t remember the exact filename. Something about primes.

I open the Skills panel. Type prime.

It searches keywords and descriptions — instant. My prime-number command shows up. I see the preview on the right: the full content.

In Claude Code I’d type /prime-number — but I had to know the exact name. Here I just type what I remember. It finds.

Scenario 2: “Design” in Headers — Structural Search

I know there’s a section called “design” somewhere in my documents. Not in the filename, not in the keywords — in a markdown header.

I type ##design.

Only ## headers are searched. Not the body, not the keywords — just the structure. Faster than full text, more precise than keyword.

"prime"       → keywords + descriptions (fast, broad)
"##design"    → headers only (fast, structural)

Two different depths. Same panel.

Scenario 3: “Exhaustive” in Body Text — Deep Search

I know I wrote “exhaustive” somewhere deep inside a document. Not in the title, not in the keywords, not in a header — somewhere in the body text.

I type ^exhaustive.

The ^ prefix triggers a full body scan. It reads through every line of every document. Slower, but finds it.

"prime"         → keyword + description  (instant)
"##design"      → headers only           (instant)
"^exhaustive"   → full body text         (slower, thorough)

Three depths. You choose based on what you know:

• Know the keyword? Just type.
• Know the section title? Use ##.
• Know it’s buried somewhere? Use ^.

Scenario 4: optionOS Trading Commands — Triple Filter

I need trading-related commands from my optionOS project specifically. Not skills, not docs — commands. Not from other projects — from optionOS.

I type /c #trading $opus.

/c          → bash commands only (not slash commands, not text)
#trading    → with "trading" keyword
$opus       → opus model only

Three filters, one line. One result.

Scenario 5: I Was Talking — “Architecture” Appeared

I’m recording with Dictation. Talking about a new feature. I say “…we need to follow the architectural design pattern…”

A popup appears at the bottom. My architectural-design command. I didn’t search for it.

MY SPEECH                     WHAT HAPPENED
─────────                     ─────────────
"...architectural design..."  → keyword "architectural" matched
                              → keyword "design" matched
                              → architectural-design.md suggested

Every document has keywords in its frontmatter. Dictation extracts words from my speech. When they match — the document surfaces.

I click it. Preview opens. I see the content, find the right section, copy it. Paste into AI. AI now has the exact context it needs.

I didn’t remember the filename. I didn’t open the Skills panel. I just talked. The right document came to me.

Scenario 6: 3 Skills, Batch Copy, One AI Session

I’m about to give AI a complex brief. It needs:

1. My architecture principles
2. My Swift patterns document
3. My database design notes

I open the Skills panel. Find the first one — Shift+click. Find the second — Shift+click. Find the third — Shift+click.

Three documents selected. Total context visible.

⌘C. Paste into AI. All three, in order, in one action.

In Claude Code I’d type /architecture, wait, type /swift-patterns, wait, type /database-design, wait. Three separate injections. No way to see them together before giving to AI.

Here: see all three previews, verify they’re the right ones, copy once, paste once. You do the orchestration.

Search Language — Complete Reference

3 document types:

/s              → slash commands
/c              → bash commands
/t              → text documents

4 search depths — each goes deeper:

prime           → keyword + description (fast, default)
#prime          → keyword only (exact match in frontmatter.keywords)
##design        → markdown headers only (structural)
^exhaustive     → full body text (slow, explicit — use when others fail)

Filters:

@Bash           → using Bash tool
@/path          → from specific path/namespace
>1k             → large documents (1000+ characters)
<500            → small documents
~w              → changed this week

Combine anything:

/c #trading $opus             → bash commands tagged "trading", opus model
/s @Bash ~w                  → slash commands using Bash, this week
#swift :personal >1k         → big swift skills in personal namespace
/t ~t recall                 → text skills modified today with "recall"