界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

Authorization

Authorization

【效率与协作】Build secure access control with RBAC, ABAC, permissions, policies, and scope-based authorization.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: Authorization slug: authorization version: 1.0.0 description: Build secure access control with RBAC, ABAC, permissions, policies, and scope-based authorization. homepage: https://clawic.com/skills/authorization metadata: clawdbot: emoji: 🔐 requires: bins: [] os: - linux - darwin - win32 displayName: Authorization

When to Use

User needs to control what actions users can perform. Agent handles permission design, role hierarchies, policy evaluation, and access control middleware.

Quick Reference

TopicFile
RBAC vs ABAC comparisonmodels.md
Implementation patternspatterns.md
Framework middlewaremiddleware.md

Core Rules

1. Auth ≠ Authorization

  • Authentication: Who you are (login, OAuth, tokens)
  • Authorization: What you can do (permissions, roles, policies)
  • Never mix concerns — auth happens BEFORE authorization

2. Principle of Least Privilege

  • Default deny — explicit grants only
  • Users get minimum permissions for their job
  • Audit permissions periodically (revoke unused)
  • Temporary elevation over permanent grants

3. Choose the Right Model

ModelBest ForComplexity
ACLSimple resource ownershipLow
RBACOrganizational hierarchiesMedium
ABACDynamic context-based rulesHigh
ReBACSocial graphs, sharingHigh

Start simple → evolve when needed.

4. Role Design Patterns

  • Roles represent jobs, not permissions
  • Max 3 inheritance levels (admin → manager → user)
  • Avoid role explosion — combine with ABAC for edge cases
  • Document role definitions (what can this role DO?)

5. Permission Naming

resource:action:scope
documents:write:own     ← Can edit own documents
documents:write:team    ← Can edit team documents
documents:delete:all    ← Can delete any document

Consistent naming prevents ambiguity.

6. Policy Evaluation Order

  1. Explicit deny → always wins
  2. Explicit allow → checked second
  3. No match → default deny
  4. Log all denials for debugging

7. Never Hardcode

// ❌ Bad — hardcoded role check
if (user.role === 'admin') { ... }

// ✅ Good — permission check
if (can(user, 'settings:update')) { ... }

Roles change. Permissions are stable.

Common Traps

  • Checking roles instead of permissions → brittle when roles change
  • OR logic in permissions → "can edit OR is admin" creates backdoors
  • Caching permissions too long → stale grants after role changes
  • Frontend-only checks → always verify server-side
  • God roles → split "admin" into specific permission sets
  • Circular inheritance → A inherits B inherits A crashes system

Security & Privacy

Data that stays local:

  • All documentation and patterns are reference material
  • No data collection or external requests

This skill does NOT:

  • Access your codebase automatically
  • Make network requests
  • Store any user data

Feedback

我的收藏