Find agent skills by outcome

- Use consistent heading hierarchy (H2 for sections

H3 for subsections)

- Use status values other than: `TODO`

`WIP`

- Notes must be scannable

rereadable

- Professional

clear

- Capture all explanations

examples

transcript_to_notes

---

## Listener-First

Time-Optimized

description: [V2] AI study assistant that transforms lectures into high-fidelity

structured notes. Optimized for AI Blaze with strict YAML schema

You are running inside AI Blaze

a browser extension. Your input is:

Education

You capture and preserve what is taught — you do not teach

reinterpret

Education

You transform learning materials (lecture transcripts

YouTube videos

costume: Patchwork leather vest

welding goggles around neck

subject_expression: Squinting against the sun

gritted teeth showing exertion.

Creative

subject_action: Hauling a heavy

salvaged turbine engine part over his shoulder.

subject_expression: Alert and scanning the horizon

eyes wide with intense focus.

subject_action: Pointing towards a distant gap in the scrap heaps

signaling a safe path forward.

costume: Layers of desert linen wraps

makeshift shoulder armor made from a rusted license plate

Harsh

directional desert sunlight

camera_angle: Low-angle medium close-up

emphasizing the scale of the junk piles behind them.

environmental_elements: Tumbleweeds made of wire

shimmering heat haze distorting the background

Echoes of the Rust Age

{

prompt: You will perform an image edit using the people from the provided photos as the main subjects. Preserve their core likeness. Place Subject 1 (male) and Subject 2 (female) as post-apocalyptic wanderers in a desert of junk. They are traversing a massive canyon formed by centuries of rusted debris. The image must be photorealistic

featuring cinematic lighting

location: A sprawling canyon formed not by rock

but by towering piles of rusted shipping containers

- Are XML tags used for containers (`<context>`

`<rules>`)?