Developer platform

Build a Vera-grade skill
on the Praetor Veritas platform.

Everything you see in Vera: the trust meter, the cross-domain answers, the autonomy ladder, the family delegation. Those are four skills built on top of one platform layer. The Skills SDK opens that same layer to any developer. Your skill surfaces in the Atelier marketplace the moment it passes review.

Request SDK access How it works

SDK ships at general availability. Preview access is by invitation. Honest framing: the examples below are forward-looking stubs that show what a third-party skill looks like in Atelier.

What you get

The substrate behind Vera.
Available to every developer.

  1. One manifest, enforced at runtime

    You declare permissions and an autonomy ceiling in a YAML manifest. The Trust Engine enforces both at the API boundary. Your skill cannot request more than it declared, and the user can revoke any permission at any time.

  2. The HITL gate is yours to use

    Any skill can mark write actions as requiring human-in-the-loop approval. The user sees the action, approves or rejects it, and the ledger records the outcome. The same gate Vera uses natively.

  3. Twelve domains, one surface

    Finance, calendar, email, health, travel, family, identity, and more. Your skill reads the data classes the user grants, in the same normalized format Vera uses, so you spend time on logic, not on parsing.

  4. Audit trail by default

    Every skill invocation is appended to the Vera audit ledger. The user can review what your skill did, when, and why. Trust is not a promise in a policy doc; it is a row in an append-only log.

Atelier preview

What a skill looks like in the marketplace.

Three stub examples showing the Atelier card format. Each declares its permissions and trust scope in a manifest.yaml. Vera enforces both at runtime.

  • Tax Advisor

    by Meridian Financial

    Finance

    Year-end tax prep, personalized to your actual transactions.

    Try "What do I need to do before tax season?"
    Trust scope Advise only, HITL gate on all write actions
    finance.readtransactions.read

  • Travel Prep

    by Waypoint Labs

    Travel

    Trip prep delivered in the order you actually need it.

    Try "I have a flight in four days. What do I need to do?"
    Trust scope Advise only
    calendar.readtravel.preferences.readidentity.read

  • Investor Update

    by Dispatch AI

    Business

    Monthly investor updates drafted and gated before you commit to send.

    Try "Draft this month's investor update."
    Trust scope Execute with approval, HITL gate required
    finance.readtransactions.reademail.send

The skills above are forward-looking stubs. They show the manifest format and Atelier card layout. Real skill execution is not yet available to third-party developers in this preview period.

The manifest

One YAML file.
The platform does the rest.

Every skill ships a manifest.yaml. You declare your permissions and autonomy ceiling. Vera validates the manifest at review time and enforces both constraints at the API boundary on every invocation. No custom auth layer. No permission-check boilerplate.

The full JSON Schema for the manifest is published at platform/schema/skill-manifest.schema.json in the Praetor Veritas platform repository. 

manifest_version: "1.0"
id: "com.yourco.my-skill"
name: "My Skill"
version: "0.1.0"

developer:
  name: "Your Company"
  url: "https://yourco.com"

category: finance

permissions:
  - finance.read
  - finance.transactions.read

trust_scope:
  max_tier: advise
  hitl_required: true
  audit_trail: true

entry:
  handler: "src.handler.handle_request"
  runtime: "python3.12"

listing:
  tagline: "One-line pitch for your Atelier card."
  sample_prompt: "Ask Vera something this skill answers."
A minimal manifest for a finance-read skill that operates in advise mode with HITL gating on any write action it proposes.