AgentSkillsCN

security

定义安全标准：密钥管理、环境变量、信任边界、威胁缓解。在处理密钥、验证输入，或实施身份认证时使用。

SKILL.md
--- frontmatter
name: security
description: Defines security standards: secret handling, environment variables, trust boundaries, threat mitigation. Use when handling secrets, validating input, or implementing auth.

Security Standards

Security is a baseline requirement, not an optional feature. All layers must be designed with a hostile environment in mind.

Core Principles

  • •Assume breach
  • •Minimize blast radius
  • •Default to deny
  • •Least privilege everywhere

Security failures are production failures.

Secrets Management

  • •Secrets must never be hardcoded
  • •Secrets must always be read from environment variables
  • •Secrets must never be logged or exposed

Rules:

  • •No secrets in source code
  • •No secrets in client bundles
  • •No secrets in error messages

Client Environment Variables

  • •Client environment variables must be exposed only when strictly necessary
  • •Every exposed variable must:
    • •Exist in the server environment
    • •Use the NEXT_PUBLIC_ prefix

Rules:

  • •Never expose secrets to the client
  • •Never infer sensitive values client-side
  • •Environment variables are configuration, not logic

Trust Boundaries

  • •All client input is untrusted
  • •Validation must occur at every boundary
  • •Never trust frontend behavior

Boundaries include:

  • •API routes
  • •Webhooks
  • •Async jobs
  • •External integrations

Principle of Least Privilege

  • •Grant the minimum permissions required
  • •Avoid broad roles or wildcard access
  • •Separate credentials per service where possible

Rules:

  • •No shared admin credentials
  • •No privilege escalation without review
  • •Access scope must be explicit

Error Exposure

  • •Never expose internal error details
  • •Stack traces must never reach clients
  • •Use generic, user-safe error messages

Rules:

  • •Detailed errors belong only in server logs
  • •Error responses must follow error-taxonomy skill

Logging & Monitoring

  • •Logs must not contain sensitive data
  • •Security-relevant events must be logged
  • •Logs must be auditable

Examples:

  • •Authentication failures
  • •Permission denials
  • •Webhook verification failures

Secure Defaults

  • •Deny by default
  • •Explicit allowlists over blocklists
  • •No implicit trust

Rules:

  • •New endpoints start closed
  • •New permissions start denied
  • •New integrations start sandboxed

Incident Mindset

  • •Assume compromise is possible
  • •Design for containment and recovery
  • •Make failures observable

Security is about reducing impact, not achieving perfection.

Non-Negotiables

  • •No secrets in code
  • •No trust in client input
  • •No internal detail leakage
  • •No security shortcuts

Security is continuous.
Every change is a potential attack surface.