VerifiedX
Protect one action free

Public starters

Start from working code.

Six real starters across TypeScript and Python that protect one risky action first, keep the workflow moving, and write real diagnostics you can inspect.

Two single-agent starters, two multi-agent starters, and two agent-plus-human starters. Each one uses the published SDKs, runs against the live API, and mirrors the downstream actions teams actually ship into systems of record.

Run doctor first

npx @verifiedx-core/sdk doctor verifiedx doctor

TypeScript · Multi-agent · OpenAI Agents

Procurement award with composed review routing.

A triage agent prepares a procurement packet, an award agent gets blocked on an unsafe supplier award, and a review agent keeps the request moving with procurement review.

  • Uses `@openai/agents` plus the native VerifiedX adapter.
  • Shows a real packet handoff plus an upstream receipt-driven review step.
README run.mjs package.json

Python · Single-agent · LangGraph

Healthcare prior auth with manual review recovery.

A single LangGraph worker queues manual review when clinicals are missing, blocks an unsafe EMR approval update, and then routes back into safe workqueue handling.

  • Model-free on purpose so the LangGraph boundary is obvious.
  • Shows an actual blocked RCM state write and a recovery path.
README run.py requirements.txt

TypeScript · Single-agent · Vercel AI SDK

High-impact support refund with fraud review fallback.

A single agent looks up the case, checks refund risk, gets blocked on an unsafe refund, and keeps the case moving with trust review.

  • Uses `generateText()` plus the native VerifiedX adapter.
  • Works as the smallest entry point before adding stream or UI wiring.
README run.mjs package.json

TypeScript · Agent + human · Claude + MCP

Compliance alert clear with human approval retry.

Claude inspects account risk over MCP, gets blocked on an unsafe alert clear, then succeeds on retry once a human analyst approval is added upstream.

  • Uses Claude Agent SDK, MCP, and the native VerifiedX adapter.
  • Shows the real composed replan path for agent plus human workflows.
README run.mjs package.json

Python · Multi-agent · OpenAI direct

High-impact support refund with composed trust review.

A triage agent prepares a refund packet, a refund agent gets blocked on an unsafe refund, and a trust review agent uses the packet plus the VerifiedX receipt to keep the case moving.

  • Uses raw OpenAI tool calling through the public Python SDK.
  • Shows composed packet passing without forcing builders into a framework.
README run.py requirements.txt

Python · Agent + human · OpenAI direct

Finance payment release with controller approval retry.

The agent inspects invoice controls, gets blocked on an unsafe payment release, then retries the same action after a human controller approval is added upstream.

  • Uses raw OpenAI tool calling through the public Python SDK.
  • Shows human approval flowing into the retry path without custom framework glue.
README run.py requirements.txt

Protect one action first

Use the same runtime you already ship. Start with one risky action, inspect the diagnostics, then widen.

Protect one action free Read docs