Package format

A mog package is a .zip archive containing a manifest, a README, and your content files. This page documents everything you need to create and publish a valid package.

Directory structure

your-package/
  mog.yaml          # Required — distribution manifest
  README.md         # Required — displayed on the listing page
  CHANGELOG.md      # Optional — version history
  content/
    SKILL.md        # Your main content (standard agentskills.io format)
    reference.md    # Additional content files
    assets/
      diagram.png   # Images are supported

The archive can have any top-level directory name — the CLI strips it during extraction. The important thing is that mog.yaml and README.md exist at the root level of that top-level directory.

mog.yaml schema

mog.yaml is the package manifest. Every field is validated on upload. The scan worker will reject the release if mog.yaml is missing or invalid.

name: acme/router-eval        # vendor/slug — see naming rules below
version: 1.0.0                # semver (major.minor.patch)
type: skill                   # skill | rule | bundle | template
description: >
  Comprehensive router evaluation patterns for React applications.
  Works with React Router v6 and TanStack Router.
targets:
  - cursor
  - claude-code
  - generic
install_map:                  # optional — override default install paths
  cursor: .cursor/skills/{slug}/
  claude-code: .claude/skills/{slug}/
entrypoint: content/SKILL.md  # optional — primary file for agent consumption
requires:                     # optional — tool requirements
  tools:
    - react-router
license: MIT                  # optional, defaults to MIT
readme: README.md             # optional — path to readme (default: README.md)

Field reference

Field	Type	Required	Description
name	string	Yes	Package identifier in vendor/slug format. Max 128 chars.
version	string	Yes	Semantic version in major.minor.patch format.
type	enum	Yes	skill, rule, bundle, template, or mcp.
description	string	Yes	Short description. 1–1024 chars. Shown in search results.
targets	array	Yes	At least one of: cursor, claude-code, codex, gemini-cli, windsurf, openclaw, generic. Use generic as the catch-all for any unlisted agent runtime.
install_map	object	No	Override the default install path per target. Use {slug} as a placeholder.
entrypoint	string	No	Path to the primary file, relative to the archive root.
requires.tools	string[]	No	Tool names this package requires. Informational only in v0.
license	string	No	SPDX license identifier. Defaults to MIT.

Naming rules

• Format: vendor/package-name
• Both parts: lowercase letters, numbers, and hyphens only
• No dots, underscores, or uppercase letters
• Examples: acme/react-testing-skill, my-org/gpt4o-rules

Package types

Type	Description
skill	A SKILL.md file following the agentskills.io specification. Teaches an agent a specific capability or workflow.
rule	Configuration rules for an agent environment (e.g. Cursor rules, Claude instructions).
bundle	A collection of multiple skills or rules packaged together.
template	A starter template or scaffold that agents can use to initialize a project.
mcp	A runnable MCP (Model Context Protocol) server. Installed by writing config to the client's MCP config (e.g. mcpServers in JSON clients, or mcp.servers for OpenClaw) rather than extracting files. See MCP Servers.

Allowed file extensions (scan)

The upload scan applies different allowlists by package type so content-only packages stay tight while MCP packages can include source and config files.

skill, rule, bundle, template

These types are content-focused. Allowed extensions include Markdown, YAML, JSON, plain text, and common images. Executable or unexpected extensions cause the scan to fail. Text-like files are scanned for accidental secrets; images are not.

Extension	Notes
.md	Scanned for secrets
.yaml, .yml	Scanned for secrets
.json	Scanned for secrets
.txt	Scanned for secrets
.png, .jpg, .svg	Images — not scanned for secrets

Extensions such as .js, .sh, .py, and .exe are not allowed for these package types.

mcp

MCP packages may include additional file types needed for runnable servers and tooling (for example TypeScript, JavaScript, Python, shell scripts, and lockfiles). See MCP Servers for the full MCP allowlist.

Install targets and paths

Target	Detection	Default install path
cursor	.cursor/ directory exists	.cursor/skills/{slug}/
claude-code	.claude/ directory exists	.claude/skills/{slug}/
codex	.codex/ directory exists	mog_modules/{slug}/
gemini-cli	.gemini/ directory exists	.gemini/skills/{slug}/
windsurf	.windsurf/ directory exists	.windsurf/skills/{slug}/
openclaw	.openclaw/ directory exists	skills/{slug}/
generic	fallback	mog_modules/{slug}/

To override the path for a specific target:

install_map:
  cursor: .cursor/rules/{slug}/     # install to rules/ instead of skills/
  claude-code: .claude/memory/{slug}/
  gemini-cli: .gemini/rules/{slug}/

MCP packages (type: mcp) do not use install_map — their install location is always a client config file. See MCP Servers for the full mcp field reference.

The lockfile (mog.lock.json)

Every mog install or mog update writes to mog.lock.json in your project root. Commit this file to version control — it makes your installs fully reproducible.

{
  "version": 1,
  "lockfileVersion": "1.0.0",
  "generatedAt": "2026-01-15T10:00:00.000Z",
  "packages": {
    "acme/router-eval": {
      "name": "acme/router-eval",
      "vendor": "acme",
      "slug": "router-eval",
      "version": "1.0.0",
      "listingId": "uuid",
      "releaseId": "uuid",
      "entitlementId": "uuid",
      "installedAt": "2026-01-15T10:00:00.000Z",
      "target": "cursor",
      "installPath": "/my-project/.cursor/skills/router-eval/",
      "archiveSha256": "a3f8c2d1...",
      "updateChannel": "minor",
      "files": [
        {
          "path": "content/SKILL.md",
          "sha256": "b1c9d2e3...",
          "size": 4096
        }
      ]
    }
  }
}

Creating a package

1. Create a directory with your content files
2. Write mog.yaml with the required fields
3. Write a helpful README.md (shown on the listing page)
4. Zip the directory: zip -r my-skill-1.0.0.zip my-skill/
5. Upload via the seller dashboard or the API
6. Wait for the scan to pass, then publish

See the seller guide for a full walkthrough including Stripe Connect setup and pricing.