external/anthropic-cybersecurity-skills/skills/implementing-endpoint-dlp-controls/SKILL.md
Implements endpoint Data Loss Prevention (DLP) controls to detect and prevent sensitive data exfiltration through email, USB, cloud storage, and printing. Use when deploying DLP agents, creating content inspection policies, or preventing unauthorized data movement from endpoints. Activates for requests involving DLP, data exfiltration prevention, content inspection, or sensitive data protection on endpoints.
npx skillsauth add seikaikyo/dash-skills implementing-endpoint-dlp-controlsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Use this skill when:
Do not use for network DLP (inline proxy-based) or cloud-only DLP (CASB).
Microsoft Purview → Data Classification → Sensitive info types
Built-in SITs for common data:
- Credit card number (PCI)
- Social Security Number (PII)
- Health records (HIPAA)
- Passport number
- Bank account number
Custom SIT example (Employee ID):
Pattern: EMP-[0-9]{6}
Confidence: High
Keywords: "employee id", "emp id", "staff number"
Microsoft Purview → Data loss prevention → Policies → Create policy
Policy Configuration:
1. Template: Financial / Medical / PII (or custom)
2. Locations: Devices (endpoint DLP)
3. Conditions:
- Content contains: Credit card numbers (min 5 instances)
- OR Content contains: SSN (min 1 instance)
4. Actions:
- Block: Prevent copy to USB, cloud, email
- Audit: Log but allow (for initial deployment)
- Notify: Show user notification with policy tip
5. User notifications:
- "This file contains sensitive data and cannot be copied to this location"
- Allow override with business justification (optional)
Monitored endpoint activities:
- Upload to cloud service (OneDrive, Dropbox, Google Drive)
- Copy to removable media (USB drives)
- Copy to network share
- Print document
- Copy to clipboard
- Access by unallowed browser (non-managed browser)
- Access by unallowed app
- Copy to Remote Desktop session
For each activity, configure:
- Audit only (log the action)
- Block with override (user can justify and proceed)
- Block (prevent action entirely)
Deploy DLP policy in "Test mode with notifications" first:
1. Policy runs in audit mode for 2-4 weeks
2. Review DLP alerts in Activity Explorer
3. Identify false positives
4. Tune SIT patterns and conditions
5. Add exclusions for legitimate workflows
6. Switch to "Turn on the policy" (enforcement)
Purview → Data loss prevention → Activity explorer
Key metrics:
- DLP policy matches per day/week
- Top matched sensitive info types
- Top users triggering DLP
- Top activities blocked (USB, cloud, email)
- Override rate (percentage of blocks overridden)
DLP incident response:
1. Review DLP alert with matched content
2. Verify sensitivity of detected data
3. Assess intent (accidental vs. intentional)
4. If intentional exfiltration → escalate to security incident
5. If accidental → educate user, refine policy
| Term | Definition | |------|-----------| | DLP | Data Loss Prevention; technology that detects and prevents unauthorized transmission of sensitive data | | SIT | Sensitive Information Type; pattern matching rules for identifying sensitive data (regex, keywords, ML classifiers) | | Policy Tip | User-facing notification explaining why an action was blocked and how to request an override | | Content Inspection | Deep inspection of file contents to identify sensitive data patterns | | Exact Data Match (EDM) | DLP matching against a specific database of known sensitive values (exact SSNs, employee records) |
development
拋棄式 HTML mockup 比稿:產出 2 到 3 個設計立場不同的變體(密度 / 版式 / 強調軸,不是換色),各附取捨說明,最後給有立場的對比結論。適用:「畫個草圖」「比較 A 版 B 版」「先看方向再做」「給我看幾種做法」。要 production 元件或設計已定案時不適用。
tools
需求不明時的意圖萃取訪談:一次一題、每題附上自己的猜測、聽出「真正想要 vs 覺得應該要」,直到能預測使用者反應(約 95% 信心)才動工。適用:需求缺少對象 / 動機 / 成功標準 / 約束,或使用者點名「訪談我」「先確認一下」「我們確定嗎」。明確自足的指示、純資訊查詢、機械性操作不適用。
development
對非平凡決策啟動新鮮 context 對抗審查(找碴不背書),在修正還便宜的時候抓出錯誤方向。適用:高風險改動(production、資安敏感邏輯、不可逆操作)、不熟的程式碼、要宣稱「這樣是安全的 / 可行的」之前。機械性操作與一行修改不適用。
testing
Reference for writing and editing agent skills well — the vocabulary and principles that make a skill predictable. Consult when authoring, reviewing, or pruning a SKILL.md.