Reference
Skills
A skill is one unit of business logic, and how it runs is its trigger. The classic skill is reactive-mail: it reads inbound mail and turns it into something structured — a verdict, a card, or a draft. A fast algorithmic rule extracts a one-time code or a tracking number; an llm skill reads a thread and drafts a reply. The same defineSkill also authors an agentic tool the AI assistant can call (in chat and over MCP) and a command button on a thread. A handful ship built in, and you add the ones your business needs — the real payoff of a stack you run yourself.
12 skills
Built in
On by default — safe to run with zero configuration. Disable any of them per deployment from Settings → Skills, which lists each skill's triggers and links any secret it still needs.
One-time codes
algorithmicPull the login code to the top of the email, ready to copy in one tap.
Password resets
algorithmicCatch password-reset emails and put the reset button front and center.
Shipping notifications
algorithmicSpot tracking emails and add a track-package button with the carrier and number.
Complaint triage
llmFlag unhappy customers, scale the urgency, and draft a brief apology for review.
Newsletters & promos
algorithmicRecognize bulk marketing mail and tuck it under a promotions tab automatically.
Spam & phishing
algorithmicCatch spoofed, unauthenticated, and phishing mail from its auth signals and links.
Spam escalation (LLM)
llmAsk the model for a second opinion on the borderline mail the rules can't call.
Unsubscribe links
algorithmicSurface a one-click unsubscribe button so leaving a list takes a single tap.
Receipts
algorithmicTurn purchase receipts into a card with the order total and number at a glance.
Calendar invites
algorithmicDetect meeting invitations and offer an add-to-calendar card with the time.
Flights
algorithmicTurn airline itineraries into a clean trip card: flight, route, and departure time.
Opt-in
These need tenant configuration (secrets, a fetch allowlist), so you add them by name with defineSkill in tinbox.config.ts. A skill names the secrets it needs with requiresEnv; set them once in Settings → Secrets (a Worker env var wins, else a sealed value) and the skill activates — until then it's gated off. They double as the reference examples for wiring a skill to your own systems.
Write your own
A skill is a small function (or an LLM prompt) plus an optional card. The built-ins are the template — copy one, point it at your own systems, and add it with defineSkill. The same umbrella authors an agentic tool the assistant can call or a command button on a thread; the trigger you give it decides how it runs.
The Writing a skill guide is the full reference: the reversible-only contract (the one design fact that governs every integration), the defineSkill signature, the SkillContext shape, a copyable handler, and how to TDD a skill offline with runEvals — no Cloudflare account needed. You can also read the built-ins on GitHub, or watch them in action on the skills showcase.