Skip to main content

Repository Model

Amplifiers is intentionally organized around a small set of clear layers.

The repository uses the Agent Skills package anatomy as its baseline for authoring skills, then adds a few repository-level layers around it.

This shape exists for one reason: the runtime skill should be easy to find, easy to attach, and easy to extend without creating parallel trees.

Canonical layers

skills/

Runtime-first packages with directly attachable instructions.

Default shape:

skills/<skill-name>/
├── SKILL.md
├── scripts/ # optional
├── references/ # optional
├── assets/ # optional
└── agents/
└── openai.yaml # optional

stacks/

Human-facing guides for recurring combinations of skills.

Default shape:

stacks/<stack-name>/
└── README.md

knowledge/

Internal repository documentation for maintainers.

docs/

The public Docusaurus site and its source files.

Why this shape works

  • skills/ stays canonical, so the runtime layer has one source of truth
  • deep material stays close to the skill that needs it
  • stacks/ describe recurring combinations without pretending to be another runtime layer
  • docs/ explains the system publicly without becoming a second authoring layer for skills
  • greenfield cleanup stays simple because obsolete structure can be removed instead of preserved

Naming rules

  • folders under skills/ and stacks/ use kebab-case
  • internal markdown files use snake_case.md
  • README.md is reserved for human-facing entry points
  • SKILL.md is reserved for runtime-first packages

Naming rules for skills

Prefer names that match the way the team thinks when attaching context to a task.

Use four practical patterns:

  1. broad specialist, such as designer
  2. specialist qualified by artifact or channel, such as email-copywriter
  3. modifier or effect, such as humanizer
  4. stack named by delivery type, such as landing-page

Greenfield rule

This repository is treated as greenfield.

Remove obsolete structure instead of preserving compatibility layers.

Static docs publishing

The docs site is authored with Docusaurus, but deployed as static output.

  • source files live under docs/
  • npm run build emits static files into docs/build/
  • GitHub Actions uploads that build artifact to GitHub Pages

So the production site is a static website, not a live Node application.