> ## Documentation Index
> Fetch the complete documentation index at: https://docs.snapgen.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Skill

> Install the snapgen-ai-models Agent Skill so Codex, Claude Code, Cursor and other coding agents choose SnapGen models, check prices and track jobs safely.

`snapgen-ai-models` is an [Agent Skill](https://agentskills.io) for coding
agents. It teaches the agent one reliable path from a request to a delivered
result: find models in the live catalog, read the schema and price before
paying, pick MCP or the CLI, submit once, keep the task id, poll to a final
status, and report the model, status, charge and outputs.

## Install

```bash theme={null}
# current project
npx skills add snapgen-ai/snapgen-devtools --skill snapgen-ai-models

# all projects for this user
npx skills add snapgen-ai/snapgen-devtools --skill snapgen-ai-models -g
```

The installer needs Node.js 22.20 or newer; the SnapGen CLI needs Node.js 20.
Reload or restart your coding client afterwards so it discovers the skill.

## Give it an execution path

The skill works through either:

* the hosted [MCP server](/integrations/mcp) (`https://api.snapgen.org/mcp`), or
* the [CLI](/integrations/cli) (`npx -y @snapgen/cli`).

Set your key as `SNAPGEN_API_KEY` or run `snapgen auth login`. Never put the key
in a prompt, a file or a repository.

## Verify without a paid task

Ask your agent:

```text theme={null}
Confirm that the snapgen-ai-models skill is loaded and say whether you will use
the SnapGen MCP tools or the SnapGen CLI. List the current image-to-video
models with their input requirements and public prices. Do not submit any job.
```

## What the skill enforces

| Rule                                | Why                                                                     |
| ----------------------------------- | ----------------------------------------------------------------------- |
| Look up model ids live              | Model ids and prices change; guessed ids fail.                          |
| Read schema and price first         | The agent can state the expected cost before spending.                  |
| Submit once with an idempotency key | A retried submission returns the same task.                             |
| Never call waiting complete         | `queued` and `processing` are not results.                              |
| Never retry blindly                 | Check the task and balance before a new attempt, which is a new charge. |
| Report a receipt                    | Model, final status, charge in USD, outputs and task id.                |

The skill never replaces the live catalog, prices or task status; it tells the
agent when and how to read them.