This repo contains installable Agent Skills for Interfaze AI. Follow these rules when working with or contributing to these skills.
- Interfaze AI is a proprietary model. Which easily work with OpenAI SDK by setting the baseURL to https://interfaze.ai/v1
- Only document how to use the API — inputs, outputs, and patterns.
- Interfaze exposes an OpenAI-compatible API. Use the Vercel AI SDK (
@ai-sdk/openai) withbaseURLpointed at the Interfaze endpoint. - The model name is
interfaze-beta. - Always require
INTERFAZE_API_KEYas an environment variable.
- Keep skill descriptions precise and trigger-oriented. The
descriptionfield is the primary routing surface. - Prefer narrow, capability-specific skills over broad catch-all skills.
- Keep examples concrete and realistic. Show actual code with Zod schemas and Vercel AI SDK calls.
- Put reusable API details and examples in
references/files, not insideSKILL.md. - Do not duplicate long API documentation inside
SKILL.md. Reference it instead. - Maintain naming consistency: folder names must match the
namefield in frontmatter.
- Use direct, practical language.
- Avoid marketing language, hype, or superlatives.
- Lead with what the user needs to do, not what the product can do.
- Prefer imperative instructions: "Use this when..." not "This skill provides..."
- Use
generateObjectfrom the Vercel AI SDK when structured output is needed. - Use
generateTextwhen plain text output is sufficient. - Define schemas with Zod. Use
.describe()on fields when the field name alone is ambiguous. - Pass images as
{ type: "image", image: "<url>" }in message content arrays. - Pass audio as
{ type: "file", data: "<url>", mediaType: "audio/mp3" }in message content arrays.
- The Interfaze API behavior is the source of truth. If a skill contradicts observed API behavior, update the skill.
- Do not invent capabilities that have not been tested against the live API.