AgentSkillsCN

privacy-guard

在代码处理前，对个人身份信息（PII）、敏感数据及未经授权的文件类型实施零容忍扫描。

SKILL.md
--- frontmatter
name: privacy-guard
description: Enforces zero-tolerance scanning for PII, secrets, and unauthorized file types before code processing.

Privacy Guard Protocol

File Scope (Strict Allowlist)

ONLY process:

  • •Code: .go, .rs, .zig, .ts, .js, .py, .c, .cpp, .h, .css, .html
  • •Build: go.mod, go.sum, build.zig*, Cargo.*, package.json, *lock*, requirements.txt, Pipfile, Makefile
  • •Config: Dockerfile, *.yaml/yml, .env.example, .gitignore, .editorconfig, .toml, .json (only if infrastructure/build config), **/skills/**/*.md (skill files)

REJECT immediately:

  • •Documents: .pdf, .docx, .doc, .rtf, .pages
  • •Data: .xls*, .csv, .numbers, user-record JSON/YAML/XML
  • •Secrets: .pem, .key, id_rsa, secrets.*

Privacy Scan (Execute Before Processing)

Detect and redact to <REDACTED>:

  • •API keys (AWS, Stripe, etc.)
  • •Database passwords
  • •Real names (non-author)
  • •Email addresses (non-dummy)
  • •Phone numbers
  • •Physical addresses
  • •Credit cards
  • •Internal IPs (192.168.x.x, 10.x.x.x) → <INTERNAL_IP>

Execution

  1. •Validate file type against allowlist
  2. •Scan for PII/secrets
  3. •Redact matches + report types found
  4. •Proceed with request OR output: 🚫 PRIVACY GUARD: File/content rejected

Never output real PII or use it in examples.