Developer quickstart
Draft your first AI workflow test.
Use a public example with Claude, Codex or your chosen assistant. No FieldSignal account is required.
Start with the replacement example.
- Download the fictional test case. It describes an urgent replacement request with unconfirmed compatibility.
- Download the test task brief. Give both files to your assistant and ask it to prepare the test and list the changes needed to run it in your tools.
- Review the output. The agent must request a compatibility check and leave the replacement order unplaced. Require a way to inspect both system states.
The result is a draft test and an integration plan. The JSON separates agent input from evaluator-only expectations; keep that separation when adapting it. It is a fictional example, not a supported API response or proof that a test has passed.
Run it in your test environment.
A developer must supply the real tool names, adapter and isolated test environment. Check that the test can detect a forbidden order even when the agent’s final response sounds correct. Keep the result marked “not run” until execution produces evidence.
Use your own workflow.
- Build an AI agent: collect records, approve rules and complete the build brief.
- Prepare for production: review results, release responsibilities and unresolved gaps.
Use only data approved for your chosen environment. Access to case records does not authorize the assistant to change business systems.
Connect FieldSignal data.
Request platform access and the private setup instructions for MCP, the CLI and FieldSignal’s Claude or Codex skills. Tell us the workflow, source systems, chosen development tools and who can approve data access.
Connection details and available data formats must be agreed before using your own FieldSignal records. The public example above does not require that setup.