Skip to main content
The Taberna Agent Skill is a packaged version of this documentation, written for the coding agent rather than for you. Install it once and the agent you already use — Claude Code, Cursor, Codex, Copilot, Gemini CLI and around forty others — writes Taberna integrations that verify webhook signatures correctly, dedupe correctly, and stop inventing endpoints that do not exist. It follows the Agent Skills open standard: a folder with a SKILL.md file, loaded progressively. Only the name and description sit in the agent’s context at rest; the body loads when a task actually mentions Taberna, and the reference files load only when a task needs that depth. Installing it costs you roughly a sentence of context until the moment it is useful.

​
What it teaches an agent

Call GET /v1/store at boot and assert the scopes you need, rather than probing an endpoint and catching the 403. The skill makes this step zero of every integration, and teaches the 401-versus-403 distinction so an agent never writes a retry loop around a missing scope.
When to reach for a catalogue order versus a composed invoice, the exact request bodies, what the 201 actually returns, and the rule that you redirect to the checkoutUrl Taberna gave you rather than building the URL yourself.
The HMAC_SHA256(secret, "{timestamp}.{rawBody}") scheme, a complete Node verifier, the framework-specific hooks for getting raw bytes, constant-time comparison, and the five-minute timestamp tolerance you have to enforce yourself. This is the single most common integration bug and the skill leads with it.
That a manual resend reuses the same X-Taberna-Delivery-Id and buys exactly one more attempt — so deduplicating on the delivery id silently drops the replays a merchant is sending precisely because something went wrong. The guard belongs in your own business state.
That there is no test mode and no refund endpoint, that every schema violation returns the same opaque {"error":"Invalid request data"}, that a malformed short id is a 400 and not a 404, that fiat is a decimal string and crypto is a smallest-unit integer string, and that there is one webhook URL per store with no event filtering.
An agent’s most expensive habit is fabricating a plausible endpoint. The skill lists the complete /v1 surface and names the things people invent: a sandbox mode, a tbrn_test_ prefix, refunds, API routes for creating products or keys, a second webhook URL, subscriptions, a storeId parameter.

​
Install

The skills CLI reads this site’s discovery index, shows you what is available, and installs it into whichever agents it finds:
This is the only method that updates cleanly — re-run it to pick up a new version.
The raw file lives at /.well-known/agent-skills/taberna/SKILL.md, and the discovery index listing it is at /.well-known/agent-skills/index.json. Point any tool that understands the standard at https://docs.taberna.io and it will find them.

​
What you get

SKILL.md is self-contained — an agent that has only that file can still build a correct integration. The reference files exist so the common path stays short: they are pulled in only when a task needs that level of detail. The full bundle, including references, is in the docs repository.
Every page on this site is also available as Markdown by appending .md to its URL — for example /orders.md — alongside /llms.txt and /llms-full.txt. The skill points at these, so an agent can always reach the authoritative page rather than working from a summary.

​
If you would rather not install anything

Instructions for AI agents is the same guidance as a page you can paste, link or point a URL-reading agent at. The skill is that content packaged so the agent loads it on its own, at the moment it becomes relevant, without you remembering to.
A skill improves the odds, it does not replace review. There is no test mode — every payment method is a live chain moving real funds, and there is no refund endpoint. Read what your agent wrote before it touches a key.