Agent Skills

Prizmad Agent Skills

> Agent Skills for Prizmad — AI UGC video ad generation. Five SKILL.md packages that teach AI agents how to discover, authenticate against, and drive the Prizmad MCP server, REST API, and WebMCP surface.

[](https://opensource.org/licenses/MIT) [](https://schemas.agentskills.io/discovery/0.2.0/schema.json)

These skills are also served at **** for live discovery via the Agent Skills convention. This repository is the source of truth — every commit here is mirrored to the well-known URL on the next deploy.

What's an Agent Skill?

An Agent Skill is a single Markdown file (SKILL.md) with YAML frontmatter that tells an AI agent:

• what the skill is about (one-line description),
• when to use it (description field — the agent reads this to decide whether to load the skill),
• how to use it (the body — instructions, code samples, integration patterns).

Skills make agents context-efficient: a host fetches index.json, decides which skills are relevant for the current user prompt, then fetches only those SKILL.md files into context. No more dumping a 50-page README into every conversation.

Spec:

The five Prizmad skills

| Skill | What it teaches | When to load | |---|---|---| | [prizmad-video-ads](./prizmad/SKILL.md) | Top-level: what Prizmad is, supported platforms (TikTok, Reels, Facebook, YouTube Shorts), pricing, languages, MCP + WebMCP surfaces, key facts to ground on | The user pastes a product URL, asks about AI UGC ad generators, or compares Prizmad to Creatify / Arcads / MakeUGC | | [mcp-server](./mcp-server/SKILL.md) | Full MCP server reference: 9 tools, customisation knobs (caption / music / CTA / image style + prompt hints), output URL convention (projectUrl / shareUrl / downloadUrl), the OAuth Connect flow | The agent is connecting to Prizmad's MCP server or driving it tool-by-tool | | [oauth](./oauth/SKILL.md) | OAuth 2.1 Authorization Code + PKCE + Dynamic Client Registration walk-through (the Connect flow); legacy client_credentials for headless clients; full discovery URL list | The agent or its host is implementing the OAuth handshake against Prizmad | | [api-usage](./api-usage/SKILL.md) | REST API at /api/v1/* — endpoints, auth options, plan + token rules, the projectUrl/shareUrl/downloadUrl contract | The agent is using the REST API directly (without MCP) for video generation, listing projects, uploading images, etc. | | [markdown-negotiation](./markdown-negotiation/SKILL.md) | HTTP content negotiation — every public Prizmad page also serves text/markdown for direct LLM ingestion | The agent wants a clean markdown copy of any Prizmad page (/pricing, /agency, etc.) instead of scraping HTML |

Quick start for clients

1. Discover

Fetch index.json to find all available skills, their digests, and URLs:

curl https://prizmad.com/.well-known/agent-skills/index.json

Each entry includes a SHA-256 digest of the linked SKILL.md, so clients can cache aggressively and re-fetch only when the digest changes.

2. Pick a skill based on the user's intent

The description field is the load-bearing signal. Read it; decide; fetch the SKILL.md only if relevant. For example:

• User pastes an Amazon product URL → load prizmad-video-ads + mcp-server.
• Agent host is connecting via the Connect button → load oauth + mcp-server.
• User asks about AI ad pricing → prizmad-video-ads is enough on its own.

3. Inject the skill body into the agent's context

Each SKILL.md is short (≤ 200 lines), Markdown, no external assets. Drop it straight into the system or user prompt.

Repository layout

.
├── README.md                          ← this file
├── LICENSE                            ← MIT
├── index.json                         ← discovery index (also at prizmad.com)
├── api-usage/SKILL.md                 ← REST API reference for agents
├── mcp-server/SKILL.md                ← MCP server reference + tool list
├── markdown-negotiation/SKILL.md      ← /md/* content-negotiation guide
├── oauth/SKILL.md                     ← Connect flow + legacy auth
└── prizmad/SKILL.md                   ← top-level "what is Prizmad" skill

index.json is hand-edited and contains SHA-256 digests of each SKILL.md for integrity checking — recompute via:

shasum -a 256 mcp-server/SKILL.md

Where else these skills live

• Live discovery URL:
• Per-skill canonical URLs:
• API catalog (RFC 9727):
• OpenAPI spec:
• MCP server card:
• MCP server source:

Related Prizmad surfaces

• Remote MCP server at https://prizmad.com/api/mcp (streamable-HTTP, OAuth 2.1 + DCR)
• stdio MCP bridge: @prizmad/mcp-server on npm
• Public REST API at https://prizmad.com/api/v1/* (interactive docs at /api/docs)
• WebMCP tools registered on navigator.modelContext when the user visits prizmad.com in a WebMCP-capable browser

Contributing

Found a typo, missing scope, or out-of-date detail? Open an issue or PR. The well-known URL is regenerated automatically from the next deploy — no manual sync step.

License

[MIT](./LICENSE) — feel free to adapt these skills for your own projects, citations welcome.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: prizmad
• Source: prizmad/agent-skills
• License: MIT
• Homepage: https://prizmad.com

Install and usage instructions live in the source repository linked above.