Standalone, copy-paste task recipes the agent (or a human) can run for common, repeatable jobs in this repo. Each prompt is a self-contained .md file with step-by-step instructions.
Structure
prompts/
├── <category>/
│ ├── <prompt-name>.mdSKILL.md VS Prompts
While both provide instructions for AI agents, they differ in purpose and how they are triggered:
- Skills (
SKILL.md): Active capabilities loaded automatically by the agent on-demand when user intent matches the skill's description. They provide technical patterns, guidelines, and domain expertise. The skill description from frontmatter is always attached within the context of the agent. (e.g.,react-aria,api-integration). - Prompts (
.md): Passive recipes invoked manually by the user or agent for specific jobs. They are linear, step-by-step task briefs. Prompt gets loaded into the context whenever explicitly invoked. (e.g.,add-page,change-branding).
When to Use
- The user names a common task (e.g. "add a new page", "change branding").
- You want a consistent, repeatable workflow instead of improvising.
- The change is mechanical enough to be scripted but still needs judgement.
- Won't worth adding into skills (e.g. too specific, too simple, etc).
Categories
| Folder | Purpose |
|---|---|
core/ | Core app setup - branding, new pages, foundational flows. |
layout/ | Sidebar / nav structure - menu groups, icons, section removal. |
ui/ | UI-level one-off tasks. |
auth/ | Auth-related changes - API integration, social login, etc. |
Authoring a New Prompt
- Drop a
<task-name>.mdinto the right category folder. - Open with a clear
# Task: <verb + object>heading. - Force input gathering first (Step 0) - never assume.
- Number steps in order; keep them imperative and verifiable.
- End with a verification step the user can confirm.
