Healthcare Agents — 51 US Healthcare Admin Specialists
A portable pack of 51 specialist AI agents and 16 standard workflows for US healthcare administration, spanning revenue cycle, quality compliance, clinical operations and more, producing appeal packets, audit checklists and workplans.
The skill mandates progressive loading (compact index before full prompts), defaults to no PHI handling, forbids unauthorized external actions (form submission, messaging, uploads), and keeps regulated decisions with named human owners — safety boundaries are explicit. Deducted for: unverified publisher identity, 'approved environment' relying solely on user assertion, and no rollback/audit mechanism.
Routing logic is self-consistent: workflow index → area/specialist index → full source prompt → output mode, with completion criteria and Completed/Partial/Blocked terminal states. Deducted for: static review cannot verify that referenced files (references/*., ../../agents/*.md) resolve; both supplied index JSON files end mid-entry, and there is no test or failure-feedback evidence, so the score cannot exceed 10.
Triggers and anti-triggers are explicit (16 workflows, 10 domains, 51 specialists), four output modes are clear, and non-fit boundaries (not clinical/legal/coding authority) are declared. Deducted for: entirely US-healthcare-administration focused with no Chinese-language support; installation depends on npx/GitHub with uncertain mainland-China reachability.
Documentation is well layered (SKILL.md router + indexes + source prompts), Apache-2.0 license is explicit, with version (v1.6.0), release notes, and uninstall/update paths. Deducted for: unclear maintenance ownership (unverified publisher), marketing-heavy claims around Operator OS and CSO protocols, and no example outputs directly in the assessed path.
Workflows define required inputs, artifact sections, red flags and handoffs, so structured administrative deliverables are plausibly producible. Deducted for: no execution evidence in a static review; output quality rests on inference from prompt design alone, and comparative benefit is unverified, so the score cannot exceed 7.
The repository provides structured indexes, coverage status tables, and citation-card status taxonomies — some auditable primary material exists. Deducted for: claims of 51/51 evaluation and a 94.18 average score are author assertions without independently reproducible test suites or third-party execution evidence; fact/inference separation is incomplete, keeping the score below 5.
- This is a static source review; no skill instructions were executed and no outputs were reproduced — confidence is low.
- Evaluation scores in the README (51/51, 94.18) are author claims and were not independently verified.
- The skill targets US healthcare administration exclusively (HIPAA, CMS, HEDIS); it does not fit Chinese healthcare contexts and offers no Chinese-language support.
- Installation depends on npx/npm and GitHub downloads, which may be unreachable from mainland-China networks.
- Despite PHI guardrails, do not enter real patient data outside an approved environment.
- The skill is decision support only; all clinical, legal, coding, and compliance decisions require a qualified human owner.
What does this skill do, and when should you use it?
Healthcare Agents is a portable prompt and SKILL.md pack providing 51 narrow specialist AI agents for US healthcare administration across ten domains: revenue cycle, quality/safety/compliance, clinical operations, payer, health IT, operations, population health, pharmacy, strategy, and emergency preparedness. It routes workflow-first: a request is matched against 16 predefined workflows (denial spike, HIPAA evidence checklist, HL7/FHIR incident, etc.) before a primary specialist is selected with supporting handoffs. Every agent supports four output modes — quick triage, workplan, audit/checklist, and artifact/template. The pack offers decision support only; it makes no final clinical, legal, coding, billing, or compliance determinations and is not a safe PHI-processing environment.
The SKILL.md acts as a self-directing router: it reads references/workflow-index. to match workflow triggers, optionally reads references/agent-index. to select the narrowest matching specialist, then loads the full agents/<slug>.md source prompt before answering. Deliverables include appeal packets, audit and readiness checklists, gap analyses, dashboard specifications, pharmacy contract scorecards, and discharge-barrier workplans. Multi-step work maintains a compact working ledger (target outcome, verified facts, required documents, pending actions, blockers, red flags) and ends with a Completed/Partial/Blocked terminal status. Workflow-first routing applies the workflow's required inputs, artifact sections, red flags, and handoff agents. A companion CLI (workup, evidence-pack, review evaluate commands) supports offline evidence packs and structured first-pass routing.
- A hospital revenue cycle team facing an 18% spike in commercial payer denials and climbing AR days needs a structured denial-spike diagnosis and appeal packet.
- A quality compliance officer must draft a HIPAA Security Rule evidence checklist for a small clinic's vendor security review.
- Clinical administration staff need to route and prepare a prior authorization appeal with the correct specialist and handoffs.
- An operations manager wants a workplan with owners, dependencies, and KPIs for ED boarding or ambulatory access backlog.
- A health IT engineer handling an HL7/FHIR interface incident needs the interoperability specialist's triage and recovery workflow.
- Pharmacy or payer teams assessing pharmacy contract performance or value-based-care downside-risk readiness need a scorecard or readiness review.
What are this skill's strengths and limitations?
- 51 narrow specialists replace a generic 'healthcare assistant', naming real mechanics: CARC/RARC, 837/835, HEDIS, Stars, 340B, HL7v2, FHIR R4.
- Workflow-first routing via compact indexes reduces irrelevant starting context for long tasks.
- Installer covers 14+ tools (Claude Code, Codex, Cursor, Windsurf, Copilot, OpenCode, Aider, etc.) with plain Markdown as the fallback.
- Ships a frozen rubric and auditable eval evidence for 51/51 agents (average latest score 94.18 — internal prompt-rubric evidence only).
- Explicit safety boundaries: no final regulated decisions, offline by default, case data loaded only in explicit modes with provenance labels.
- No live search, payer portal connectors, private PHI ingestion, or MCP adapters — all excluded from the default runtime.
- Only denial-spike is a full exemplar workflow; the other 15 are standard packs without golden-artifact scoring.
- Eval scores are against the repo's own rubric — not external certification or benchmarks; most public citation cards are not verified pinpoint citations.
- Regulations, code sets, and payer policies change; verify against current primary sources before operational use.
- Administration support only — not a substitute for final clinical, legal, coding, billing, audit, or compliance determinations and not a HIPAA compliance tool.
How do you install this skill?
Fastest: run npx --yes healthcare-agents install, or target a specific tool, e.g. npx --yes healthcare-agents install --claude (Claude Code subagents into ~/.claude/agents), --claude-skills, --codex, --cursor, --copilot, --opencode, or --all for all known targets; add --dry-run to preview. Codex plugin route: git clone https://github.com/ajhcs/healthcare-agents.git then bash scripts/install-codex-plugin.sh. Also available: npx doctor to inspect install targets and collisions, --force to update, uninstall --all to remove. Apache-2.0 licensed.
How do you use this skill?
After installing, invoke with natural phrasing — no internal skill name required: "Use the Healthcare Agents plugin for a denial spike.", "Use the quality department for a survey readiness checklist.", "Route this prior authorization appeal." The skill matches one of the 16 workflows (denial spike, clean claim decline, underpayment review, prior auth appeal, discharge barrier, HIPAA evidence checklist, survey readiness, patient safety RCA2, ED boarding, ambulatory access backlog, downside-risk readiness, HEDIS/Stars gap closure, HL7/FHIR incident, clinical dashboard spec, pharmacy contract scorecard, emergency preparedness readiness) or picks a specialist by department/area, then answers in one of four output modes. Do not supply PHI outside an approved environment with minimum necessary controls.
How does this skill compare with similar options?
The README contrasts the pack with generic healthcare prompting (e.g., "Follow HIPAA", "Improve revenue", "Check quality"): generic prompts cannot distinguish Privacy Rule from Security Rule, CARC/RARC from 837/835, or HEDIS from Stars, whereas this pack supplies named, source-aware specialist prompts per domain.