zhongwei/gh-jnlei-claude-tools-plugins-skills-skill-developer

Fork 0

Files

Zhongwei Li 13e0eb6ac7 Initial commit

2025-11-30 08:28:15 +08:00

8.5 KiB

Raw Permalink Blame History

skill-rules.json - Complete Reference

Complete schema and configuration reference for .claude/skills/skill-rules.json.

File Location
Complete TypeScript Schema
Field Guide
Example: Guardrail Skill
Example: Domain Skill
Validation

File Location

Path: .claude/skills/skill-rules.json

This JSON file defines all skills and their trigger conditions for the auto-activation system.

Complete TypeScript Schema

interface SkillRules {
    version: string;
    skills: Record<string, SkillRule>;
}

interface SkillRule {
    type: 'guardrail' | 'domain';
    enforcement: 'block' | 'suggest' | 'warn';
    priority: 'critical' | 'high' | 'medium' | 'low';

    promptTriggers?: {
        keywords?: string[];
        intentPatterns?: string[];  // Regex strings
    };

    fileTriggers?: {
        pathPatterns: string[];     // Glob patterns
        pathExclusions?: string[];  // Glob patterns
        contentPatterns?: string[]; // Regex strings
        createOnly?: boolean;       // Only trigger on file creation
    };

    blockMessage?: string;  // For guardrails, {file_path} placeholder

    skipConditions?: {
        sessionSkillUsed?: boolean;      // Skip if used in session
        fileMarkers?: string[];          // e.g., ["@skip-validation"]
        envOverride?: string;            // e.g., "SKIP_DB_VERIFICATION"
    };
}

Field Guide

Top Level

Field	Type	Required	Description
`version`	string	Yes	Schema version (currently "1.0")
`skills`	object	Yes	Map of skill name → SkillRule

SkillRule Fields

Field	Type	Required	Description
`type`	string	Yes	"guardrail" (enforced) or "domain" (advisory)
`enforcement`	string	Yes	"block" (PreToolUse), "suggest" (UserPromptSubmit), or "warn"
`priority`	string	Yes	"critical", "high", "medium", or "low"
`promptTriggers`	object	Optional	Triggers for UserPromptSubmit hook
`fileTriggers`	object	Optional	Triggers for PreToolUse hook
`blockMessage`	string	Optional*	Required if enforcement="block". Use `{file_path}` placeholder
`skipConditions`	object	Optional	Escape hatches and session tracking

*Required for guardrails

promptTriggers Fields

Field	Type	Required	Description
`keywords`	string[]	Optional	Exact substring matches (case-insensitive)
`intentPatterns`	string[]	Optional	Regex patterns for intent detection

fileTriggers Fields

Field	Type	Required	Description
`pathPatterns`	string[]	Yes*	Glob patterns for file paths
`pathExclusions`	string[]	Optional	Glob patterns to exclude (e.g., test files)
`contentPatterns`	string[]	Optional	Regex patterns to match file content
`createOnly`	boolean	Optional	Only trigger when creating new files

*Required if fileTriggers is present

skipConditions Fields

Field	Type	Required	Description
`sessionSkillUsed`	boolean	Optional	Skip if skill already used this session
`fileMarkers`	string[]	Optional	Skip if file contains comment marker
`envOverride`	string	Optional	Environment variable name to disable skill

Example: Guardrail Skill

Complete example of a blocking guardrail skill with all features:

{
  "database-verification": {
    "type": "guardrail",
    "enforcement": "block",
    "priority": "critical",

    "promptTriggers": {
      "keywords": [
        "prisma",
        "database",
        "table",
        "column",
        "schema",
        "query",
        "migration"
      ],
      "intentPatterns": [
        "(add|create|implement).*?(user|login|auth|tracking|feature)",
        "(modify|update|change).*?(table|column|schema|field)",
        "database.*?(change|update|modify|migration)"
      ]
    },

    "fileTriggers": {
      "pathPatterns": [
        "**/schema.prisma",
        "**/migrations/**/*.sql",
        "database/src/**/*.ts",
        "form/src/**/*.ts",
        "email/src/**/*.ts",
        "users/src/**/*.ts",
        "projects/src/**/*.ts",
        "utilities/src/**/*.ts"
      ],
      "pathExclusions": [
        "**/*.test.ts",
        "**/*.spec.ts"
      ],
      "contentPatterns": [
        "import.*[Pp]risma",
        "PrismaService",
        "prisma\\.",
        "\\.findMany\\(",
        "\\.findUnique\\(",
        "\\.findFirst\\(",
        "\\.create\\(",
        "\\.createMany\\(",
        "\\.update\\(",
        "\\.updateMany\\(",
        "\\.upsert\\(",
        "\\.delete\\(",
        "\\.deleteMany\\("
      ]
    },

    "blockMessage": "⚠️ BLOCKED - Database Operation Detected\n\n📋 REQUIRED ACTION:\n1. Use Skill tool: 'database-verification'\n2. Verify ALL table and column names against schema\n3. Check database structure with DESCRIBE commands\n4. Then retry this edit\n\nReason: Prevent column name errors in Prisma queries\nFile: {file_path}\n\n💡 TIP: Add '// @skip-validation' comment to skip future checks",

    "skipConditions": {
      "sessionSkillUsed": true,
      "fileMarkers": [
        "@skip-validation"
      ],
      "envOverride": "SKIP_DB_VERIFICATION"
    }
  }
}

Key Points for Guardrails

type: Must be "guardrail"
enforcement: Must be "block"
priority: Usually "critical" or "high"
blockMessage: Required, clear actionable steps
skipConditions: Session tracking prevents repeated nagging
fileTriggers: Usually has both path and content patterns
contentPatterns: Catch actual usage of technology

Example: Domain Skill

Complete example of a suggestion-based domain skill:

{
  "project-catalog-developer": {
    "type": "domain",
    "enforcement": "suggest",
    "priority": "high",

    "promptTriggers": {
      "keywords": [
        "layout",
        "layout system",
        "grid",
        "grid layout",
        "toolbar",
        "column",
        "cell editor",
        "cell renderer",
        "submission",
        "submissions",
        "blog dashboard",
        "datagrid",
        "data grid",
        "CustomToolbar",
        "GridLayoutDialog",
        "useGridLayout",
        "auto-save",
        "column order",
        "column width",
        "filter",
        "sort"
      ],
      "intentPatterns": [
        "(how does|how do|explain|what is|describe).*?(layout|grid|toolbar|column|submission|catalog)",
        "(add|create|modify|change).*?(toolbar|column|cell|editor|renderer)",
        "blog dashboard.*?"
      ]
    },

    "fileTriggers": {
      "pathPatterns": [
        "frontend/src/features/submissions/**/*.tsx",
        "frontend/src/features/submissions/**/*.ts"
      ],
      "pathExclusions": [
        "**/*.test.tsx",
        "**/*.test.ts"
      ]
    }
  }
}

Key Points for Domain Skills

type: Must be "domain"
enforcement: Usually "suggest"
priority: "high" or "medium"
blockMessage: Not needed (doesn't block)
skipConditions: Optional (less critical)
promptTriggers: Usually has extensive keywords
fileTriggers: May have only path patterns (content less important)

Validation

Check JSON Syntax

cat .claude/skills/skill-rules.json | jq .

If valid, jq will pretty-print the JSON. If invalid, it will show the error.

Common JSON Errors

Trailing comma:

{
  "keywords": ["one", "two",]  // ❌ Trailing comma
}

Missing quotes:

{
  type: "guardrail"  // ❌ Missing quotes on key
}

Single quotes (invalid JSON):

{
  'type': 'guardrail'  // ❌ Must use double quotes
}

Validation Checklist

JSON syntax valid (use jq)
All skill names match SKILL.md filenames
Guardrails have blockMessage
Block messages use {file_path} placeholder
Intent patterns are valid regex (test on regex101.com)
File path patterns use correct glob syntax
Content patterns escape special characters
Priority matches enforcement level
No duplicate skill names

Related Files:

SKILL.md - Main skill guide
TRIGGER_TYPES.md - Complete trigger documentation
TROUBLESHOOTING.md - Debugging configuration issues

8.5 KiB Raw Permalink Blame History