Prototype · Agent Fabric Console

Multi-agent control plane

Live view of every Pause-Health.ai agent currently registered on a (mocked) MuleSoft Agent Fabric: the Agentforce Inbound Lead Generation, Prospecting & Nurture, Qualification, and Engagement agents that bracket the patient lifecycle, Agentforce intake, the Assessment Agent that deterministically scores validated instruments (MRS, Greene, PHQ-9, ISI) into an intake severity, the Benefits & Coverage Verification (EBV) Agent that runs a synthetic eligibility check before routing, the Anthropic Claude-backed Care Router, the Care Plan Agent that instantiates a template-sourced menopause care plan and summarizes progress with live Claude (the second live-Claude agent, with a graceful scripted fallback like the Care Router), the Appointment Scheduling Agent that books the recommended MSCP visit against a synthetic provider calendar and hands it to engagement for reminders, the Referral Management Agent that triages intake + routing signals into cosign-gated outbound specialist referrals (generalizing the Care Router's behavioral-health handoff), the Member Service / Billing Agent that answers claim-sourced billing & coverage self-service questions and routes out-of-scope requests to a human, the Prior Authorization Agent (the heaviest, deliberately-last workflow) that assembles a clinician-gated, documentation-complete PA and never autonomously submits it, the Care Gap Closure Agent that proactively detects Data-360-grounded, clinical-measure-sourced preventive-care gaps and drafts consent-aware outreach for engagement, the Medication Adherence Agent that proactively tracks HRT/SSRI adherence + refill timing and drafts nudge-only refill reminders (never an autonomous refill) for engagement, the Clinical Summary Agent that composes the other agents' outputs into a patient-friendly after-visit summary and a clinician handoff with live Claude (the third live-Claude agent, with the same scripted fallback), grounding every summary in the source records the context was assembled from, the SDOH Screening Agent (whole-person care) that screens a patient for health-related social needs with the validated CMS AHC-HRSN core-domain tool, escalates the interpersonal-safety red flag to a human social worker, and drafts consent-gated community-resource referrals that are never an autonomous enrollment, the Patient Education & Health Coaching Agent that turns the intake, care-plan, and care-gap signals into a deterministically-selected, evidence-sourced menopause/midlife education curriculum and coaches the patient with live Claude (the fourth live-Claude agent, with the same scripted fallback), staying strictly within general education (no diagnosis, dosing, or individualized medical advice) with consent-gated outreach, the Remote Patient Monitoring & Symptom-Trend Tracking Agent that ingests longitudinal symptom/vital readings, deterministically detects per-metric trends against a synthetic monitored-metrics catalog, and routes worsening or red-flag trends to a clinician for review without ever taking an autonomous clinical action, the Population Health & Risk Stratification Agent that reasons over a whole patient panel at once, deterministically scores each patient with a transparent, additive risk model (no protected-class attributes) into a low/rising/high tier, and builds a prioritized outreach worklist for a human care manager — never an autonomous care decision, the Clinical Trials & Research Matching Agent that deterministically matches a single patient against a synthetic study catalog using structured eligibility criteria, ranks the matching studies with per-criterion explanations, and drafts a research-consent-gated outreach that never auto-enrolls (informed consent + a human required), the Language Access & Health Equity Agent that determines a limited-English-proficiency patient's preferred language, deterministically decides whether a qualified medical interpreter is needed and of which modality, checks approved in-language materials, and flags equity gaps — using a qualified medical interpreter only (never a family / ad-hoc / machine interpreter), never machine-translating clinical consent, and escalating to a human coordinator when no qualified interpreter is available, the HEDIS & Quality Reporting Agent that deterministically rolls up a whole panel against a defined HEDIS measure catalog into per-measure numerator / denominator / catalog-sourced exclusions / compliance rate for value-based-care contracts, and assembles a submission package that ALWAYS requires human quality-team approval (never autonomously filed, and never inflated by an ad-hoc / unlisted denominator exclusion), the Advance Care Planning Agent that uses perimenopause / menopause as a midlife touchpoint to surface which advance directives are on file (living will, DPOA-HC; POLST only for serious-illness patients), flags missing / stale / language-access gaps, and drafts a consent-gated conversation prompt for the care team — every directive on file traces to the catalog + an approved source, every directive change is clinician + patient sign-off gated (never autonomously applied), and for an LEP patient with no interpreter plan the active prompt is withheld until the Language Access agent has arranged a qualified interpreter (a safe answer, not a block), the Care Team & Case Management Agent that assembles the multi-disciplinary team around a single high-need patient (PCP, MSCP, cardiology, endocrinology, bone-health, pelvic-floor PT, behavioral health), assigns a case manager by a stable-hash pick from a synthetic pool, and emits a shared team snapshot — every role traces to the catalog, every roster change requires case-manager sign-off (never autonomously applied), and a legitimate team must include a PCP anchor, the Discharge & Transitions of Care Agent that closes the loop back to primary care after a hospitalization / ED visit — deterministically reconciling the discharge medication list (added / removed / dose-changed, each tracing to an approved medication source and every change clinician-signoff gated), booking (or handing off to Scheduling for) the follow-up appointment as a real slot (never a text recommendation — the load-bearing 30-day-readmission guard), pulling encounter-reason red-flag warning signs, emitting the teach-back checklist, and assembling the PCP handoff summary, the Grievance & Appeals Agent that runs the intake half of the regulated grievance-and-appeals process — deterministically classifying a member complaint or coverage-denial appeal, routing to the correct human queue (member-services / clinical-review / compliance), stamping a catalog-sourced regulatory deadline (never silently extended past the maximum), and handing the receiving queue a PHI-safe routing summary (structured only, never free-text PHI) — the agent never resolves a case on its own, the Quality-Measure Attribution Agent that pairs with the HEDIS agent to decide whose panel each patient counts on — deterministically attributing each patient to a provider / clinic / VBC contract under a catalog-sourced methodology (plurality-of-visits, PCP-of-record, prospective Medicare Advantage, contract-defined window), honoring the contract's exclusion terms so an excluded patient doesn't pollute the scorecard, applying a documented tie-break chain (most-recent-visit-wins → provider-ref-lexical-ascending) instead of a gameable coin-flip, and rolling up per-provider counts so downstream HEDIS scoring lands on the correct denominator, the Complex Care Management Agent that runs the reimbursable time-tracking piece of a Medicare CCM program — deterministically confirming eligibility (2+ catalog-sourced chronic conditions, Medicare-eligible age, coverage flag, consent), tracking per-activity minutes against the CCM activity catalog, mapping the total to the CPT ladder (99490 → 99491 → 99487 → 99489), and assembling a billing package for human quality-team review (never autonomously submitted to CMS, never inflated by phantom minutes), the Claims Adjudication Assistant Agent that runs the first-pass payer-side pipeline — deterministically applying catalog edits (NCCI-PTP, LCD/NCD, benefit limits, prior-auth linkage, duplicates, network, timely-filing), classifying each claim as clean-pay / pend / deny-drafted with a specific catalog reason code, and routing non-clean items to a human (never autonomously finalizing a denial — every denial letter needs an adjudicator cosign), the Formulary & Drug Utilization Review Agent that runs the first-pass DUR pipeline — deterministically checking tier, step-therapy (honored with documented prior-therapy), quantity limits, and drug-drug interactions, classifying as preferred-approved / pend with a catalog reason code, and routing non-preferred decisions to a clinician cosign (never an autonomous formulary exception override), the Fraud, Waste & Abuse Detection Agent that screens claims and prior-auths against catalog patterns (unbundling, upcoding, duplicate billing, quantity outliers, impossible-day billing, phantom services), classifies severity, and routes to the SIU for HUMAN review — never autonomously denies a claim, opens an investigation, or freezes payment, and never scores on protected-class attributes (a documented compliance failure in real payer FWA systems), the Clinical Trial Payments & Stipends Agent that pairs with Clinical Trials Matching to handle the reimbursable payments side — deterministically computing per-visit stipend + travel reimbursement against IRB-approved schedules, verifying research-payment consent (45 CFR 46), and routing non-standard payments to the study coordinator for cosign (never autonomously deviates from an IRB schedule), the Utilization Review Agent (MCG/InterQual analog) that runs the pre-service medical-necessity screen against catalog criteria sets for proposed procedures / inpatient admissions, classifies as approves-meets-criteria / pend-for-clinical-review / require-peer-to-peer / blocked-non-covered with a specific reason code, routes non-approved cases to a clinical reviewer or peer-to-peer with a catalog-sourced SLA deadline (standard 72h / urgent 24h / concurrent-review 24h), and never autonomously denies (every non-approved decision requires clinician cosign — a Medicare Advantage / state UR-agent due-process requirement), the Care Coordination Handoff Agent (Joint-Commission-NPSG-2 SBAR) that handles any cross-setting patient transition (hospital → SNF, SNF → home, home → hospice, ED → PCP, PCP → specialist, PCP → behavioral health) — deterministically assembling the SBAR (situation, background, assessment, recommendation), verifying the receiving clinician's credentialing status, and confirming transfer consent for transitions that share PHI with a new setting; classifies as handoff-accepted / pend-sbar-incomplete / blocked-clinician-not-credentialed / blocked-no-consent, and NEVER autonomously accepts on behalf of the receiving clinician (distinct from Transitions of Care, which is post-discharge hospital→home + med reconciliation), the Adverse Event Reporting Agent (FDA MedWatch / VAERS analog) that runs the pharmacovigilance / device-safety reporting pipeline — deterministically classifying each drug ADR, vaccine reaction, device malfunction, medication error, or therapeutic failure into the MedWatch (3500 / 3500A) or VAERS channel, computing the 21-CFR-314.80 seriousness tier (non-serious / serious / life-threatening / death) from caller-provided outcome flags, verifying reporter identity attestation, and drafting for regulatory-team cosign — NEVER autonomously files to the FDA (21 CFR 314.80 mandatory reporting has sponsor / manufacturer / clinician liability) and NEVER drafts on an unverified reporter (FDA reporting requires an attested reporter), the Data-Sharing / TEFCA Interoperability Agent that classifies each cross-organization PHI exchange (TEFCA QHIN / Carequality / CommonWell / Direct Secure Messaging) by exchange purpose (treatment / payment / operations / patient-request / public-health / research), verifies the counterparty is a Trusted Exchange Framework participant, applies the patient's data-sharing consent scopes from the Consent agent, and classifies as release-authorized / pend-purpose-verification / blocked-non-catalog-purpose / blocked-participant-unverified / blocked-consent-required-non-tpo — NEVER autonomously releases PHI for a non-TPO purpose without consent (HIPAA §164.506 boundary — TPO doesn't need consent, everything else does) and NEVER releases to an unverified counterparty (45 CFR 171 / TEFCA Common Agreement), the Consent & Preferences Management Agent — the authoritative, cross-cutting consent ledger + communication-preference store the other agents' consent-before-outreach / consent-before-referral / consent-to-monitor gates defer to, deterministically deciding whether a patient may be contacted for a scope over a channel at a time while honoring revocations/expiries immediately and never overriding a scope, the Pause MCP server, the MCP Bridge that lets fabric agents call external MCP servers, the MuleSoft Process API, the Provider Credentialing & Directory Agent that gates every referral / scheduling attempt at the network boundary (verifying credentials against approved sources, blocking referrals to expired / incomplete / sanctioned providers, and refusing to return stale directory records past the No-Surprises-Act 90-day accuracy window as authoritative — the ghost-network fix), and — on a strictly PHI-separated commercial plane — the Pipeline Management, Account Management, and Provider Contracting & VBC Terms agents (the last classifies provider-network contracts, computes the VBC quality-gate + spend-benchmark drift against a catalog methodology, and drafts term-change proposals that a human account owner must sign off on — never autonomously commits a contract-term change). Every A2A handoff and tool call lands here as a trace span so you can govern, monitor, and audit the multi-agent system in one place.