Przejdź do treści
Biblioteka skilli
Bezpieczeństwo · Planowanie

dlp

/dlp

Kieruje zapytania o Data Loss Prevention: klasyfikuje je na warstwy endpoint, sieć i chmura (USB, e-mail, CASB, odkrywanie danych) i deleguje do odpowiedniego agenta technologii.

Repozytorium
domain-expert ↗
Licencja
MIT
Ostatnia zmiana repozytorium
23.09.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
2
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 23.09.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Kieruje zapytania o Data Loss Prevention: klasyfikuje je na warstwy endpoint, sieć i chmura (USB, e-mail, CASB, odkrywanie danych) i deleguje do odpowiedniego agenta technologii.

Wyzwalacze z opisu

DLP\data loss prevention\data exfiltration\sensitive data leak\data classification\content inspection\USB block\email DLP\
Oryginalny opis (EN)

Expert routing agent for Data Loss Prevention. Classifies DLP requests across endpoint, network, and cloud enforcement layers, then delegates to the appropriate technology agent. WHEN: \"DLP\", \"data loss prevention\", \"data exfiltration\", \"sensitive data leak\", \"data classification\", \"content inspection\", \"USB block\", \"email DLP\", \"cloud DLP\", \"CASB\", \"data discovery\". Do NOT use for vendor-specific questions -- use the `purview-dlp`, `symantec-dlp`, `forcepoint`, `digital-guardian`, or `cyberhaven` skill.

Jak działa

  1. How to Approach Tasks
  2. Technology Routing
  3. DLP Concepts Reference
  4. Core DLP Knowledge
  5. Technology Agents
  6. Reference Files
Sekcje z nagłówków SKILL.md, w oryginale · 218 linii, 1 398 słów
Podgląd SKILL.md
---
name: dlp
description: "Expert routing agent for Data Loss Prevention. Classifies DLP requests across endpoint, network, and cloud enforcement layers, then delegates to the appropriate technology agent. WHEN: \"DLP\", \"data loss prevention\", \"data exfiltration\", \"sensitive data leak\", \"data classification\", \"content inspection\", \"USB block\", \"email DLP\", \"cloud DLP\", \"CASB\", \"data discovery\". Do NOT use for vendor-specific questions -- use the `purview-dlp`, `symantec-dlp`, `forcepoint`, `digital-guardian`, or `cyberhaven` skill."
license: MIT
---

# Data Loss Prevention (DLP)

This skill covers Data Loss Prevention: DLP concepts, architecture, and the major DLP platforms. It helps prevent unauthorized exfiltration or exposure of sensitive data across endpoint, network, and cloud channels.

## How to Approach Tasks

When you receive a DLP request:

1. **Identify the technology** — Determine which DLP platform is in use (Purview, Forcepoint, Symantec, Digital Guardian, Cyberhaven, or technology-agnostic).

2. **Classify the request type:**
   - **Architecture/Design** — Load `references/concepts.md` for DLP fundamentals
   - **Policy configuration** — Read the technology-specific sibling skill
   - **Detection tuning** — Understand data types, then read the sibling skill
   - **Incident response** — Understand enforcement channel, then read the sibling skill
   - **Data discovery** — Identify scope (endpoint, network, cloud), then read the sibling skill

3. **Load context** — Read `references/concepts.md` for general DLP concepts, or read a sibling skill for platform-specific work.

4. **Cross-reference** — Read the appropriate sibling skill using the decision tree below.

5. **Recommend** — Provide actionable guidance with specific policy examples.

## Technology Routing

### Microsoft Purview DLP
**Read `purview-dlp` when:**
- Microsoft 365 environment (Exchange, SharePoint, OneDrive, Teams)
- Purview compliance portal or Unified labeling
- Endpoint DLP on Windows or macOS (M365-managed)
- Sensitivity labels, MIP, AIP
- Copilot/AI Hub data leak prevention
- Adaptive Protection or Insider Risk Management integration
- Keywords: "Purview", "M365 DLP", "MIP", "sensitivity labels", "endpoint DLP", "compliance portal", "EDM", "exact data match", "trainable classifier", "Activity Explorer", "Content Explorer"

### Forcepoint DLP
**Read `forcepoint` when:**
- Forcepoint ONE or Forcepoint DLP platform
- Behavioral analytics / risk-adaptive protection
- Dynamic data protection based on user risk scores
- Forcepoint Web Security or Email Security DLP integration
- Keywords: "Forcepoint", "dynamic data protection", "risk-adaptive", "Forcepoint ONE", "behavioral analytics DLP"

### Symantec DLP (Broadcom)
**Read `symantec-dlp` when:**
- Symantec DLP / Broadcom DLP environment
- Network Monitor, Network Prevent (email/web)
- Endpoint Discover, Endpoint Prevent
- Indexed Document Matching (IDM) or Exact Data Matching (EDM)
- Vector Machine Learning (VML)
- Keywords: "Symantec DLP", "Broadcom DLP", "Vontu", "Network Monitor", "Endpoint Prevent", "IDM", "VML", "Detection Server"

### Digital Guardian (Fortra)
**Read `digital-guardian` when:**
- Digital Guardian agent (kernel-level)
- Fortra Digital Guardian platform
- ARC (Analytics & Reporting Cloud)
- Agent-based data visibility across all data movement
- Keywords: "Digital Guardian", "Fortra DLP", "DG agent", "ARC", "kernel-level DLP"

### Cyberhaven
**Read `cyberhaven` when:**
- Cyberhaven platform
- Data lineage tracking
- Behavioral data flow analysis
- Generative AI data protection (ChatGPT, Copilot leakage)
- Keywords: "Cyberhaven", "data lineage", "data flow tracking", "behavioral DLP"

## DLP Concepts Reference

Load `references/concepts.md` for general DLP architecture questions, vendor-neutral policy design, detection method selection, and data classification framework design.

## Core DLP Knowledge

### What DLP Protects Against

Data Loss Prevention prevents three primary scenarios:

- **Exfiltration** — Intentional theft by malicious insider or compromised account
- **Negligent exposure** — Accidental sharing, misconfigured permissions, wrong recipient
- **Inadvertent leak** — Auto-sync to personal cloud, copy to unmanaged USB, AI prompt injection

### Data Classification Tiers

| Tier | Label | Examples | Default Treatment |
|---|---|---|---|
| 1 — Public | Public | Press releases, public docs | No restrictions |
| 2 — Internal | Internal | General business data | Internal sharing only |
| 3 — Confidential | Confidential | Client data, contracts, strategy | Need-to-know, no external sharing without approval |
| 4 — Restricted | Highly Confidential | PII, PHI, PCI data, trade secrets | Strict controls, encryption required, audit all access |

### Detection Methods

**Regex / Pattern Matching**
- Fastest, lowest compute
- Best for structured data (SSN: `\d{3}-\d{2}-\d{4}`, credit card patterns, passport numbers)
- High false positive rate without context
- Use with keyword boosters (near "social security", near "card number")

**Exact Data Matching (EDM / Fingerprinting)**
- Hash fingerprints of actual sensitive records (employee database, customer PII list)
- Near-zero false positives — only triggers on actual data values
- Requires periodic re-fingerprinting as data changes
- Best for: protecting known PII datasets, customer lists, source code files

**Document Fingerprinting**
- Creates fingerprint of a template document
- Detects filled-out versions of that template (W-9s, NDAs, contracts)
- Useful for: HR forms, financial templates, legal agreements

**ML / Trainable Classifiers**
- Trained on labeled document corpora
- Classifies document type by content semantics, not just patterns
- Built-in classifiers: financial statements, HR resumes, source code, medical records
- Lower precision than EDM but catches novel sensitive content
- Best for: unstructured data where exact patterns are unknown

**OCR (Image/Screenshot DLP)**
- Extracts text from images, screenshots, PDFs
- Applies pattern/classifier detection to extracted text
- Significant compute cost — apply selectively
- Best for: detecting screenshots of confidential data, image-based document exfil

### Enforcement Channels

```
Endpoint DLP
├── Clipboard monitoring (copy/paste to unauthorized app)
├── USB/removable media blocking or auditing
├── Print / print to PDF blocking
├── Screen capture restrictions
├── Cloud sync client monitoring (OneDrive personal, Dropbox, Google Drive)
└── Browser upload blocking

Network DLP
├── Email (SMTP) — inspect outbound email attachments and body
├── Web (HTTP/HTTPS) — inspect web uploads (requires SSL inspection)
├── File transfer protocols (FTP, SFTP)
└── Instant messaging / collaboration tools

Cloud DLP (CASB)
├── Shadow IT discovery (unsanctioned SaaS usage)
├── API-based inspection of cloud storage (Box, Dropbox, Google Drive)
├── Inline proxy or reverse proxy for SaaS
└── Cloud-to-cloud data movement monitoring
```

### DLP Policy Design Workflow

```
1. Identify data assets
   └── What sensitive data exists? Where does it live?
       └── Run data discovery scan (endpoint, file shares, cloud)

2. Classify sensitivity
   └── Apply classification labels (manual, automatic, or recommended)
   └── Define what makes each tier sensitive (regulatory requirements, business value)

3. Define policies
   └── For each data tier: Who can access? What can they do? What is blocked?
   └── Map to enforcement channels: endpoint + network + cloud

4. Start in audit/monitor mode
   └── Collect baseline — understand normal data flows before blocking
   └── Tune false positives before enabling enforcement

5. Enforce with graduated response
   └── Low risk: log only
   └── Medium risk: alert + user notification ("this looks sensitive")
   └── High risk: block + require justification
   └── Critical: block + alert security team + trigger incident

6. Review and iterate
   └── Activity Explorer / SIEM review
   └── Tune policies based on false positives/negatives
   └── Update fingerprint/EDM databases as data changes
```

### Regulatory Drivers for DLP

| Regulation | Key Data Type | DLP Requirement |
|---|---|---|
| HIPAA | PHI (patient health info) | Prevent unauthorized disclosure; audit all access |
| PCI DSS | Cardholder data (PAN, CVV) | Restrict to CDE; detect and block exfiltration |
| GDPR | EU personal data | Data minimization; detect unauthorized transfers outside EU |
| CCPA | California resident PI | Know where data is; prevent unauthorized sale/disclosure |
| SOX | Financial records | Integrity controls; audit financial data access |
| ITAR/EAR | Technical data, export-controlled | Prevent transfer to unauthorized persons/countries |

### Common DLP Anti-Patterns

**Starting with block mode** — Always begin in audit/monitor mode. Blocking before understanding normal data flows causes business disruption and loss of user trust.

**Over-broad regex without context** — A SSN regex alone matches phone numbers, ZIP codes, invoice numbers. Add keyword context, minimum occurrence thresholds, and confidence scoring.

**Ignoring endpoint DLP** — Network DLP misses encrypted traffic, VPN users, and remote workers who are off-network. Endpoint DLP is essential for comprehensive coverage.

**Fingerprinting stale data** — EDM databases become outdated as records change. Stale fingerprints miss new sensitive records or generate stale matches. Automate re-fingerprinting on a schedule.

**No user education** — DLP without user awareness increases shadow IT as users route around controls. Pair enforcement with clear user-facing messages explaining the policy and how to get exceptions.

**Single-channel focus** — Organizations often deploy email DLP and consider themselves protected. Attackers and careless users use cloud sync, USB, and web uploads. Enforce across all channels.

## Technology Agents

Read these sibling skills for platform-specific work:

- `purview-dlp` — Microsoft Purview DLP (M365, endpoint, Copilot)
- `forcepoint` — Forcepoint DLP (risk-adaptive, behavioral analytics)
- `symantec-dlp` — Symantec/Broadcom DLP (IDM, EDM, VML, Network Monitor)
- `digital-guardian` — Digital Guardian/Fortra (kernel-level endpoint DLP)
- `cyberhaven` — Cyberhaven (data lineage, behavioral DLP, AI protection)

## Reference Files

- `references/concepts.md` — DLP fundamentals: classification tiers, detection methods, policy design patterns, regulatory mapping, CASB integration
Pliki w paczce (2)
  • SKILL.md 11 KB
  • references/concepts.md 15 KB

Wersje

W bazie nie ma innych skilli o nazwie dlp.

Migawka bazy z 13.11.2026 · wersja paczki 2b314e3c80b3 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100implementing-endpoint-dlp-controls
    Wdraża kontrole DLP na stacjach roboczych, aby wykrywać i blokować wyciek wrażliwych danych przez e-mail, USB, chmurę i druk, z agentami DLP i politykami inspekcji treści.
  • 90dlp-alert-triage
    Weryfikuje alerty DLP, odróżnia zwykły proces biznesowy od wycieku danych i dobiera proporcjonalną reakcję z zachowaniem prywatności.
  • 95implementing-data-loss-prevention-with-microsoft-purview
    Wdraża polityki DLP w Microsoft Purview przez PowerShell i Graph API: etykiety poufności, typy danych wrażliwych, reguły na endpointach i monitorowanie.
  • 100insider-threat-program-data-loss-prevention-dlp
    Wykrywa i ogranicza ryzyko wewnętrzne przy użyciu narzędzi zapobiegania utracie danych i monitoringu zachowań w organizacji.
  • 100purview-dlp-integration
    Integruje Microsoft Purview DLP i audyt z agentem Microsoft Agent 365, blokując wrażliwe dane (karty, SSN, PII) przed modelem i audytując odpowiedzi.
  • 90data-classification-to-dlp-protocol
    Prowadzi przepływ od wykrywania wrażliwych danych przez klasyfikację etykietami Microsoft Purview po polityki DLP i monitoring przyjęcia etykiet w Microsoft 365 i Power Platform.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.