Skip to main content
Skills are scoped bundles of context, prompts, and examples you can drop into an AI coding agent (Claude Code, Cursor, others) so it operates Verial correctly without you re-teaching the vocabulary every session. A good skill tells the agent what environments, simulators, benchmarks, criteria, and runs mean, which endpoints to call, and how to read verdicts and evidence.
A first-party Verial skills package is on the roadmap. Until then, this page describes the patterns we recommend so you can hand-author a skill for your team’s agent.

​
What a Verial skill should cover

  • Vocabulary. Environments, simulators, benchmarks, tasks, criteria, verification, runs, task runs, criterion runs, playgrounds, sandboxes, interactions. See the Core Concepts pages.
  • Auth model. Organization API keys for authoring, Solver keys for running published benchmarks, run-scoped bearer tokens for driving rollouts. See Authentication and Solver Keys.
  • The two entry points. Internal POST /benchmark-runs for authoring teams, public POST /v1/benchmark-runs for external Solvers. See Runs.
  • The six criterion check types. fhir-resource-state, hl7-structural, portal-state-match, sftp-file-present, voice-transcript, x12-response. See Criteria.
  • How to read results. Verdicts, scores, axes, per-criterion evidence. See Run Results.

​
Starter prompt

Point your coding agent at this file, Core Concepts > Environments, and Running a Benchmark as always-loaded context. A minimal seed: 
You operate on Verial, a healthcare AI agent benchmarking platform. Verial concepts:
environments compose simulators + datasets; benchmarks group tasks with typed
criteria; runs produce per-criterion scores via the verification engine. Auth:
organization API keys (vk_*) for authoring; Solver keys (vrl_slv_*) for driving
published benchmarks; run-scoped bearers (vrl_run_*) for rollout calls.

Reference: https://docs.verial.ai
MCP tools: see /mcp/tools

​
Alternatives

If you are on an MCP-capable agent, MCP does most of the work a skill would — tools are self-describing and the agent can discover them at runtime. Use a skill when the agent needs richer vocabulary and workflow context than tool schemas alone carry.

​
Next Steps

MCP Setup

Connect an MCP-capable agent to Verial.

Guided Onboarding

Have an agent run its first evaluation end to end.