convez.ai
MCP ServerSegurançaAvançado
markl-a

Phantom Secure Connector

markl-a/phantom-secure-connector

Desidentificação PHI + conformidade + red/blue team para Claude e Cursor

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que combina desidentificação de dados sensíveis (PHI), verificação de conformidade regulatória (HIPAA/GDPR) e simulação de segurança red/blue team. Protege dados de saúde em trânsito e em repouso, com suporte nativo a identificadores taiwanese (RG, NHI, MRN) e ocidentais (SSN, DOB, email).

Instalação

$npx convez-ai markl-a/phantom-secure-connector

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 3 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Sanitizar dados de pacientes antes de processar em pipelines IA
  • Validar conformidade HIPAA/GDPR em datasets médicos e financeiros
  • Simular ataques de prompt injection para verificar robustez de agentes

Veredito

Alpha maduro com Tier 1 pronto (desidentificação reversível + checker de conformidade + MCP bridge). Solução específica para compliance em saúde/finança; exige Python stdlib e integração manual com Claude Desktop. Roadmap claro até Q3 2026 com LLM-augmented PHI detection e validação em dados reais.

README

A tradução para o português deste README ainda está na fila.

phantom-secure-connector

CI

PHI 去識別化 + 合規檢查 + MCP bridge,加紅藍隊模擬橋接 — phantom-mesh 安全套件(紅藍隊目前橋接 phantom-secops,Tier 2 內建),跨產業招聘對齊 (資安 / 金融 / 醫材 / AI safety),目標是把三到四家 vendor 的能力收斂到一處。

status: alpha · Tier 1 license: Apache-2.0 phantom-mesh ecosystem

30-second demo

docs/demo.cast — asciinema recording of the PHI redactor tokenising a synthetic patient line (DOB / TW mobile / email).

# play in a terminal (requires asciinema)
asciinema play docs/demo.cast

# or view the captured text without any tooling:
cat docs/demo.cast | jq -r '.[] | select(.[1]=="o") | .[2]'

Self-hosted on purpose — no upload to asciinema.org, no third-party tracking.

一句話 niche

"Sensitive data comes in safely, trusted tools go out safely, behaviour is continuously red/blue-team verified — for phantom-mesh."

Presidio 只做 de-id;NeMo Guardrails 只做 LLM 輸出守門;Garak 只做 red-team; HIPAA 合規 scanner 是另一坨 SaaS。**phantom-secure-connector 把 de-id + 合規

  • MCP bridge 組合在一起,並橋接 phantom-secops 的 red-team(輸出守門為 Tier 2 路線)的 phantom 套件** — 中文 (台灣身分證 / 健保 卡 / 病歷號) 與西方識別符 (SSN / DOB / MRN / Email) 一次處理。

Status (2026-05-22)

  • ✅ Tier 1 shipped (stdlib-only,本機可跑 pytest):
    • phi_redactor/ — TW 身分證 + NHI + MRN + SSN + Email + DOB regex, 可逆 / 不可逆 mode + reversible mapping。
    • compliance_checker/ — CSV/JSON 對 HIPAA / GDPR 規則檔掃描。
    • mcp_bridge/ — Claude Desktop / Cursor 可掛 3 個 phantom tool stub。
  • 🟡 Tier 2 next: LLM-augmented PHI catches(regex 抓不到的邊界 case via phantom-mesh provider trait)、secops_simulator/ 從 stub 升級(現只是 介面,要接 phantom-secops 真實 prompt-injection harness)。
  • 🟡 Tier 3 (M2-M3, ~2026-07): AHI 真實資料 vs Nassi et al. 2021 baseline 驗證、MCP server polish 對齊最新 spec、HealthKit / Garmin Connect ingest pipeline (redact → encrypt → FTS5)。

30-second quickstart

git clone https://github.com/markl-a/phantom-secure-connector
cd phantom-secure-connector
python3 -m pytest -v       # all tests, stdlib-only

PHI redactor

from phi_redactor.redactor import redact

text = "張三 1990/01/15 出生, MRN-A123456, alice@example.com SSN 123-45-6789"
clean, mapping = redact(text, mode="replace")
# clean   → "張三 [DOB_1] 出生, [MRN_1], [EMAIL_1] [SSN_1]"
# mapping → reversible map back to originals

Compliance checker

python3 -m compliance_checker.checker --standard hipaa path/to/data.csv

MCP bridge

See mcp_bridge/README.md for Claude Desktop config.

Architecture (within phantom-mesh ecosystem)

phantom-secure-connector 是 P4 加密為先 的資料平面守門員 — 任何進 phantom-mesh memory 的東西先過 PHI redactor;任何離開的工具呼叫過 MCP bridge。(時序異常偵測 anomaly_detector 已移至 phantom-companion 的健康數據平面。)

External data (HealthKit / CSV / chat)
       ↓
   phi_redactor  ←──  TW 身分證 / NHI / MRN / SSN / DOB / Email
       ↓
   compliance_checker  ←── HIPAA / GDPR rule files
       ↓
   phantom-mesh FTS5  (encrypted at rest)
       ↓
   mcp_bridge  ←──  Claude Desktop / Cursor (露 redact_phi / fts5_search / event_capture)
       ↑
   secops_simulator  ──→  red/blue prompt-injection 持續驗證

Pillars served: P4 (加密為先,主要)、P1 (跨平台 — stdlib-only 確 保 Mac / Windows / Linux 一致行為)、P2 (分身連線 — MCP bridge 是 phantom-mesh 與外部 IDE 的安全橋)。

Target users (recruiter / co-builder angle)

跨產業最廣的一個 phantom-mesh satellite:

  • 趨勢 Trend Micro — security AI / SOC tooling、prompt-injection 防護
  • 中信 / 國泰 / 富邦 — 金融合規 + PHI/PII de-id 管線
  • 醫材公司 / 工研院生醫 — HIPAA + de-id + AHI 異常偵測背景
  • Anthropic — MCP ecosystem + AI safety / red-team
  • Co-builders: 想要 self-hosted Presidio + Garak + NeMo Guardrails 三合一 的 AI infra 工程師。

NOT doing(邊界):非醫材(supportive,not diagnostic);非網路防火牆 / IDS (資料/應用層而已);非 SOC2 證書(輔助稽核,不取代)。

Roadmap (per master plan)

3-bullet:

  1. M2 — LLM-augmented PHI(provider trait 接 phantom-mesh)、secops_simulator 從 stub 升級到真實 prompt-injection harness。
  2. M3 — AHI 真實資料驗證、MCP polish、HealthKit/Garmin ingest。
  3. Post-M3 — SOC2 audit-ready report 自動生成、企業版加密 KMS。

Sources & credit

  • ~/Documents/GitHub/hailmary/phantom-secops/ — 紅藍隊模擬器(referenced, 未複製)
  • AHI Detection v2 research — markl-a/ahi-cv5-report
  • Medical RAG v1.1.0 — PHI de-id 架構(public 版已 sanitize)

License

Apache-2.0. © 2026 Mark Lai (markl-a). See LICENSE.

Ver o original no GitHub

Bom para

Relacionados