Connect Claude to a bounded ITMS tax workflow.
Give Claude or Claude Code the public documentation and Prophit.ai skill, then add only the scoped sandbox credentials and operations required for the proof of concept.
1 add the public skill2 select a sandbox workflow3 provide scoped credentials4 inspect response + evidence5 stop at human authorityNever paste production secrets into a prompt or project file.Three steps from documentation to a safe first call.
The same pattern works in Claude.ai projects and Claude Code. The interface for attaching instructions differs, but the authority boundary does not.
- 01
Add the skill and authoritative documentation
Download the Prophit.ai Claude skill, attach it to the Claude project or repository workflow, and point Claude to the current API quickstart, schemas, errors, and rate-limit documentation. The live documentation wins if a starter file becomes stale.
- 02
Choose one sandbox operation and scope the key
Begin with a read-only or otherwise bounded operation. Keep API keys in the harness secret store or environment, not in chat, source control, skill files, command output, or logs.
- 03
Review the request, result, and audit trail
Confirm the organization, location, endpoint, input facts, response, evidence, and errors. Do not expand into filing, payment, approval, or production actions without the separate authority required for those actions.
Watch the complete AI agent path.
The walkthrough covers supported surfaces, deterministic and AI-assisted modes, downloadable harness packs, and the three-step authority checklist.
Need help choosing the first bounded workflow?
Bring the use case, data source, target endpoint, environment, and review owner. We’ll map the smallest safe proof of concept.