Skill Packages

Amplifiers follows the Agent Skills package direction used across the ecosystem, including OpenAI Codex skills and Anthropic's Claude skill guidance: one folder per skill, with the runtime entrypoint and optional support material colocated.

That is the core discipline behind this repository: each skill is a small, self-contained package instead of a loose collection of prompts and notes.

Default shape

my-skill/
├── SKILL.md          # required: instructions + metadata
├── scripts/          # optional: executable code
├── references/       # optional: documentation
├── assets/           # optional: templates, resources
└── agents/
    └── openai.yaml   # optional: OpenAI/Codex metadata

Only SKILL.md is required.

What each part does

Path	Purpose	Required
SKILL.md	Main runtime instructions and metadata	yes
references/	Longer documentation loaded only when needed	no
scripts/	Executable helpers for the skill	no
assets/	Templates, packaged resources, static inputs	no
agents/openai.yaml	Optional package metadata for OpenAI/Codex tooling	no

Current repository direction

In Amplifiers:

• runtime behavior lives in skills/<name>/SKILL.md
• deeper material lives in skills/<name>/references/
• executable helpers belong in skills/<name>/scripts/
• packaged resources belong in skills/<name>/assets/
• optional OpenAI metadata belongs in skills/<name>/agents/openai.yaml

Repository note

Repository-level adapters such as .agents/ and .claude/ are not the same thing as the internal anatomy of a skill package.

They exist only to help each runtime discover or install the canonical skills authored under skills/.

Examples from this repository

• skills/designer/
• skills/sales-copywriter/
• skills/react-architect/
• skills/dopamine-driven-copywritter/

Rule

Do not create a second top-level authoring layer when the material clearly belongs to one skill package.