Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计通过

guardrailguardrail 搜索

Agent Skill

guardrail 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

196

周安装

8

GitHub Stars

8

下载量

63
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:guardrail(guardrail 搜索)
来源仓库:https://github.com/phrazzld/claude-config
仓库路径:skills/guardrail
安装命令:
npx skills add https://github.com/phrazzld/claude-config --skill guardrail
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/phrazzld/claude-config --skill guardrail

简介

用于查找、检索和筛选与安全和合规相关的信息与策略。

  • 适合在内容审核、风险识别和政策匹配等场景中使用。
  • 通过 GitHub 安装并适用于 Codex、Claude 等宿主环境。
  • 使用前需确认权限,避免触发未授权的数据访问。
  • guardrail 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

/guardrail

Generate custom lint rules that enforce architectural decisions at edit time.

Philosophy

Lint rules are the highest-leverage codification target. They're cheaper than hooks (no custom Python), more durable than CLAUDE.md (automated, not advisory), and work in CI too (not just Claude Code). A lint rule catches violations the moment code is written — and fast-feedback.py surfaces the error immediately so Claude self-corrects.

When to Use

  • Import boundaries ("all DB access through repository layer")
  • API conventions ("routes must start with /api/v1")
  • Deprecated pattern blocking ("no direct fetch, use apiClient")
  • Auth enforcement ("handlers must call requireAuth")
  • Naming conventions that go beyond basic linting

Workflow

Phase 1: Accept Pattern

Parse the input. It can be:

  • Natural language: "all database queries must go through the repository layer"
  • Code example: "this import is wrong: import {db} from './db'"
  • Discovery mode: scan codebase for architectural invariants (when invoked by /tune-repo)

Clarify the constraint:

  • What EXACTLY should be flagged? (imports, function calls, patterns)
  • What's the fix? (alternative import, wrapper function)
  • Are there exceptions? (test files, migrations, the repository itself)

Phase 2: Choose Engine

CriterionESLintast-grep
LanguageJS/TS onlyAny (Python, Go, Rust, etc.)
FixableYes (auto-fix)Yes (rewrite)
TestingRuleTester built-inYAML snapshot tests
ConfigFlat config pluginsgconfig.yml
SpeedFastVery fast

Default: ESLint for JS/TS projects, ast-grep for everything else. If --engine is specified, use that.

Phase 3: Generate Rule

Read the reference docs for the chosen engine:

  • ESLint: references/eslint-rule-anatomy.md
  • ast-grep: references/ast-grep-rule-anatomy.md

Read the appropriate template:

  • ESLint: templates/eslint-rule.js + templates/eslint-rule-test.js
  • ast-grep: templates/ast-grep-rule.yml

Generate:

  1. Rule implementation with clear error message and fix suggestion
  2. Rule metadata (docs URL, fixable, schema)
  3. Test cases (valid AND invalid examples from the actual codebase)

Phase 4: Test

ESLint:

# Run RuleTester
node guardrails/rules/<rule-name>.test.js
# Or if project uses a test runner:
npx vitest run guardrails/rules/<rule-name>.test.js

ast-grep:

sg scan --config guardrails/sgconfig.yml --test

Also verify against the real codebase:

# ESLint: run rule on entire project, expect 0 or known violations
npx eslint --no-warn-ignored --rule 'guardrails/<rule-name>: error' .
# ast-grep: scan project
sg scan --config guardrails/sgconfig.yml

Phase 5: Install

Create the guardrails/ directory structure if it doesn't exist:

guardrails/
  README.md              # Catalog of all custom rules
  index.js               # ESLint local plugin barrel (JS/TS projects)
  sgconfig.yml           # ast-grep config (if non-JS rules exist)
  rules/
    <rule-name>.js       # ESLint rule implementation
    <rule-name>.test.js  # ESLint RuleTester
    <rule-name>.yml      # ast-grep rule

ESLint integration (flat config, zero npm dependencies):

// guardrails/index.js
import noDirectDbImport from "./rules/no-direct-db-import.js";

export default {
  rules: {
    "no-direct-db-import": noDirectDbImport,
  },
};
// eslint.config.js — add to existing config
import guardrails from "./guardrails/index.js";

export default [
  // ... existing config
  {
    plugins: { guardrails },
    rules: {
      "guardrails/no-direct-db-import": "error",
    },
  },
];

ast-grep integration:

# guardrails/sgconfig.yml
ruleDirs:
  - rules

Phase 6: Document

Update guardrails/README.md with:

## <rule-name>

**Engine:** ESLint | ast-grep
**Pattern:** <what it enforces>
**Rationale:** <why — link ADR if exists>
**Auto-fix:** yes | no
**Exceptions:** <files/patterns excluded>

Output

GUARDRAIL CREATED:
- Rule: guardrails/rules/<name>.<ext>
- Test: guardrails/rules/<name>.test.<ext>
- Engine: ESLint | ast-grep
- Violations found: N (in current codebase)
- Auto-fixable: yes | no
- Config updated: eslint.config.js | guardrails/sgconfig.yml

Anti-Patterns

  • Rules that fire on >20% of files (too broad, probably wrong constraint)
  • Rules without tests (defeats the purpose)
  • Rules without clear error messages (Claude can't self-correct from "error")
  • Duplicating built-in ESLint/Ruff rules (check first)
  • Over-specific rules that match one file (use CLAUDE.md instead)

Integration

Consumed byHow
fast-feedback.pyRuns eslint <file> and sg scan <file> on every edit
/codify-learningRoutes "lint rule" target here
/doneRoutes "lint rule" target here
/tune-repoDiscovers patterns, recommends /guardrail invocations
/check-qualityAudits guardrails/ completeness
CI (GitHub Actions)Standard eslint. or sg scan in workflow

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

33.68%
按下载量换算21

Claude

31.38%
按下载量换算20

Cursor

20.07%
按下载量换算13

Gemini CLI

10.17%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills