Skip to main content
coding_precheck runs the documentation precheck for a medical-coding encounter before CPT coding. It is part of the Nanonets Coding Engine tool group. The full precheck system prompt and the appointment-type template rules (ported from Fleming getPreCheckPrompt / ProcessVisitForPreChecks) are baked into the tool. The agent only supplies the dynamic encounter inputs — there is no need to paste a raw ai_call body.

​
Inputs

  • encounter_context (required): the full encounter note / documentation to validate, including insurance and closed-date details. May wrap the target encounter in <current_encounter> tags.
  • appointment_type (required): selects the matching documentation template.
  • encounter_duration (optional): falls back to whatever is present in the note context.
  • encounter_provider (optional): used for the sign-off / provider-match check.

​
Output

The precheck evaluates every rule and returns all blocking reasons in reasons[] (it no longer stops at the first failure), so a non-codeable encounter reports every problem consistently across runs instead of whichever single reason the model happened to pick. reason is the combined string (all reasons joined) for consumers that read a single escalation string. On a pass, reasons is empty and reason is "". Insurance eligibility states are distinguished: the precheck fails only on confirmed ineligible/inactive (“Insurance is not eligible for the visit”) or self-pay (“Insurance is Self-Pay”). An “Unverified” / unchecked eligibility does not block — it is surfaced as a non-blocking entry in warnings[] (“Insurance eligibility is unverified”) so the agent/biller is aware without escalating on it. The tool parses the {"status":"pass|fail","reasons":[...]} shape as well as the legacy single reason and {"passed":bool,...} shapes (combined and de-duped). If none parse, it returns passed=false with reason "could not parse precheck response" and includes the raw response.

​
Human-readable text

In the task feed the tool renders a ## Precheck table (result PASSED/FAILED, raw status) and a ## Reasons list with every failing reason. The full result struct stays in structuredContent.

​
Agent Instruction Guidance

If passed=false, stop the workflow and set the final escalation_reason to the precheck reason. Do not write back to the EHR.

​
Notes

  • LLM tool. Billed per call; the static prompt is cached where the provider supports it.
  • default_enabled is false (opt-in).