Skills & Media
Reusable prompt snippets, reference skills for agents, and the media catalog.
Skills
A skill is a prompt snippet managed from the dashboard and reusable anywhere — a tidy way to centralize repeated instructions (answer style, project rules, output formats) without scattering them across clients.
Manage them from the Skills page: name, description, and prompt content. The API endpoints are simple — GET /v1/skills, POST /v1/skills, PATCH /v1/skills/:id, DELETE /v1/skills/:id.
Instead of typing a skill in, you can upload a SKILL.md: its frontmatter name and description and its body fill the create form, so you can review the prompt before saving. This is the same format skills are served in, so a skill copied from one install can be uploaded to another.
A skill can be used three ways:
- Switch it on — the gateway appends its prompt to the system prompt of every inference request, for every API key, chain, and model. New skills start switched off.
- Pick it on an API key — on a key's Skills tab, choose the skills appended only to requests made with that key (
skill_idsonPATCH /v1/keys/:id). They are added on top of the switched-on ones, and a change applies from the next request. - Copy its URL — each skill is served as a Markdown document at
/skills/<id>/SKILL.mdthat an agent can read on demand.
Injected skills are appended after the client's own system prompt, in creation order, each at most once.
Skill URLs are public
A skill document needs no API key, so agents can fetch it. The id in the URL is random, but anyone who has the link can read the prompt — keep secrets out of skills.
Reference skills
Besides your own skills, the Skills page shows reference skills: ready-made guides that explain to an agent how to use Tera Router — which endpoints to call, what authentication looks like, what request bodies to send.
| Reference skill | Covers |
|---|---|
| Tera Router | Setup guide and index of the others |
| Chat | /v1/chat/completions, /v1/messages, /v1/responses |
| Image Generation | /v1/images/generations |
| Text-to-Speech | /v1/audio/speech |
| Speech-to-Text | /v1/audio/transcriptions |
| Embeddings | /v1/embeddings |
| Web Search | /v1/search |
| Web Fetch | /v1/web/fetch |
How to use them: click Copy on the skill you want, then paste its URL into your agent (Claude Code, Cursor, etc). The agent will read the guide and know how to talk to your router. Each guide fills in the address it was requested from, so the examples match your router.
Click the eye icon on any skill — reference or custom — to read its SKILL.md rendered as formatted Markdown in a dialog, exactly as an agent receives it.
Media catalog
The Media page groups providers by non-chat capability: embeddings, image, text-to-speech, speech-to-text, web search, and web fetch. It is built from the same provider catalog as the Providers page (GET /v1/media), so every card there can be connected and routed to. Each card shows whether the provider is connected and links to its detail page.