AGPM

Discovery

How AGPM discovers artifacts in source repositories.

Overview

AGPM automatically discovers artifacts by scanning source repositories for specific file patterns and directory structures.

Discovery Formats

AGPM discovers artifacts in three formats:

claude-marketplace

Repositories with a .claude-plugin/marketplace.json file:

{
  "plugins": [
    {
      "name": "plugin-name",
      "source": "./",
      "skills": ["./skills/pdf", "./skills/docx"]
    }
  ]
}

Each plugin's source directory should contain a skills/ subdirectory with individual skills.

claude-plugin

Repositories with a .claude-plugin/plugin.json file:

{
  "name": "plugin-name",
  "skills": "./custom/skills/"
}

This format allows specifying a custom skills directory path.

simple

Repositories with a top-level skills/ directory containing skill subdirectories with metadata.json:

repo/└── skills/    ├── pdf/    │   └── metadata.json    └── web-search/        └── metadata.json

Discovery Process

When you run agpm source discover or agpm add:

  1. Repository is cloned to ~/.agpm/repos/
  2. AGPM checks for .claude-plugin/marketplace.json
  3. If not found, checks for .claude-plugin/plugin.json
  4. If not found, checks for skills/ directory
  5. Scans for metadata files
  6. Parses artifact metadata

See Also