3 moxno
privacyscrubber-guard Skill
通过 @privacyscrubber/mcp-server 为 AI 编程智能体(Claude Code、Cursor、Codex、Gemini CLI、Windsurf)提供的接入前零信任 PII、密钥及受监管数据清洗技能。在本地内存中拦截文件、shell 命令和提示词,内置 30 个行业专用配置模板,并提供 SHA-256 合规回执。
安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。
技能指令原文(SKILL.md)
PrivacyScrubber Guard — Agent Skill
Enforces local in-RAM PII and secret sanitization before AI agents read files, run terminal commands, apply patches, or send context to external LLM APIs.
Quick Setup
claude mcp add privacyscrubber -- npx -y @privacyscrubber/mcp-server
Or via Python (uvx):
claude mcp add privacyscrubber -- uvx privacyscrubber-mcp
Core Agent Workflow
- Sanitize Text & Prompts (
scrub_text): Mask PII, API keys, HIPAA PHI, PCI-DSS PAN/IBAN, and corporate secrets into deterministic surrogate tokens ([EMAIL_1],[OPENAI_KEY_1],[SSN_1]) in local RAM. - Guard File Reads & Shell Commands (
guard_read_file,guard_exec,guard_apply_patch): Automatically redact sensitive literals from local files and command outputs before they enter the agent context window. - Restore Cleartext Locally (
reveal_text): Re-hydrate surrogate tokens from volatile session memory into final local outputs without exposing cleartext to external LLM providers. - Compliance Attestation (
generate_audit_receipt): Emit a deterministic SHA-256 compliance receipt mapping detected entities to GDPR, HIPAA, SOC 2, PCI-DSS, and NIST frameworks.