agentleFS
hollandkevinthollandkevint/data-product-operator/skills/healthcare-data-reconciliation/SKILL.md

healthcare-data-reconciliation

Skill5 starsChanged 2 months agoMITView on GitHub

Investigate disagreements between healthcare source records and downstream outputs. Use for claims/RCM totals, HL7/FHIR record pulls, mappings, clinical-result routing or patient-operations state. Produce an evidence trace and discriminating checks without assuming a root cause or authorizing data changes.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use approved schemas, aggregate summaries and synthetic cases.

    line 11
  2. Distinctive:

    Ask who owns source interpretation and who can accept the downstream result.

    line 17
  3. Distinctive:

    Follow source capture, extraction, delivery, identity matching, mapping, transformation, display and downstream action.

    line 22
  4. Distinctive:

    Never treat a successful transport acknowledgment as proof of downstream acceptance, payment or clinical follow-up.

    line 26
  5. Distinctive:

    Do not call the first plausible explanation the root cause.

    line 45
  6. Distinctive:

    Do not resend patient messages or change routing rules while investigating without authorization.

    line 53
  7. Distinctive:

    Test a historical failure, an unaffected case and a boundary case in an approved environment before widening use.

    line 60
  8. Distinctive:

    Keep containment, correction and prevention separate.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Healthcare Data Reconciliation

Trace a discrepancy from source to intended use and identify the next check that separates plausible explanations. Use approved schemas, aggregate summaries and synthetic cases. Keep record-level review inside the authorized environment.

Bound the disagreement

Record expected and observed behavior, affected period/population, grain, versions and consequence. Distinguish event, receipt, processing and discovery time. A later snapshot may legitimately differ. Ask who owns source interpretation and who can accept the downstream result. Separate confirmed impact from possible harm. Follow the organization's incident process if patient safety may be affected.

Trace the path

Follow source capture, extraction, delivery, identity matching, mapping, transformation, display and downstream action. At each transition record input/output identifiers, version, inclusion rule, rejected records and the available evidence. Find the last point of agreement. Compare an affected example with a relevant unaffected example. Compare rows and distinct entities as well as totals; equal totals can conceal wrong-patient links or offsetting errors. Never treat a successful transport acknowledgment as proof of downstream acceptance, payment or clinical follow-up.

Select the relevant branch

WorkflowChecks that change the conclusion
Claims analysis / RCMClaim versus line grain, correction chain, voids/reversals, payer and adjudication stage, mature comparison period; reconcile billed, allowed, paid and patient responsibility separately
Patient record pullAuthorized patient scope, requested filters, server-supported behavior, pagination, resources omitted by permissions, status and correction handling
HL7/FHIR mappingSending-system identity, code system/version, units, null semantics, repeated values, preliminary/final/corrected status and resource provenance
Clinical result routingSource result, receipt, correct patient and clinician, review queue, acknowledgment and follow-up; technical delivery alone does not close the workflow
Scheduling / referralsCurrent state and state history, cancellation/rebooking, duplicate delivery, appointment linkage, contact permission and send-time recheck
Customer support AISource authority and freshness, account/patient isolation, unsupported answers, human escalation and whether any action occurred

Pin the relevant profile, payer feed, measure or interface contract. Do not invent universal HL7 acknowledgment or payer-processing semantics.

Maintain competing explanations

For each candidate explanation, record supporting evidence, contradictory evidence and a check that could reject it. Distinguish source absence, extraction loss, access filtering, transformation error and changed consumer definitions. Do not call the first plausible explanation the root cause. A fix that passes one fixture proves only the behavior exercised. Separate original-failure reproduction from verification of the proposed correction.

Synthetic case

An interface reports successful receipt, but a laboratory result is absent from a clinician's queue. Possible explanations include identity mismatch, provider routing, status filtering, delayed processing or a later workflow failure. The next check follows the same result through each stage using approved internal evidence. Do not resend patient messages or change routing rules while investigating without authorization. Closure requires evidence of the intended follow-up, not only a green interface status.

Return a reconciliation brief

Include the discrepancy, timeline, source-to-output trace, competing explanations and next discriminating checks. For each proposed correction, identify the owner, affected scope, acceptance cases, side effects and rollback or recovery limit. Test a historical failure, an unaffected case and a boundary case in an approved environment before widening use. Plan recurrence observation over relevant exposure, with a balancing check for wrongly excluded or duplicated work. Fewer tickets alone cannot prove prevention. Do not invent a universal number of cycles or a success threshold. Keep containment, correction and prevention separate. State what remains unresolved.

Sources and limits

Reviewed September 10, 2026. The checks above are proposed workflow adaptations.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.