Component Quality Validator

Validates that generated skills, agents, and hooks are production-ready before use.

Quality Philosophy

A component is only valuable if it:

1. Works correctly - Executes without errors
2. Is discoverable - Can be found when needed (semantic matching)
3. Is actionable - Provides clear, specific guidance
4. Is maintainable - Easy to update and extend
5. Is efficient - Doesn't waste tokens or time

SKILL Quality Criteria

Required Structure (30%)

---
name: kebab-case-name        # REQUIRED: Valid kebab-case
description: |               # REQUIRED: Multi-line with triggers
  What this skill does.
  Triggers: keyword1, keyword2
version: 1.0.0               # REQUIRED: Semantic version
---

# Skill Title                # REQUIRED

## Workflow                  # REQUIRED: Step-by-step
## Code Templates            # REQUIRED: Copy-paste ready
## Validation                # REQUIRED: Success criteria

Quality Checks

/^[a-z][a-z0-9-]*[a-z0-9]$/

## Workflow

## Steps

## Validation

Validation Script

#!/bin/bash
# validate-skill.sh <skill-path>

SKILL_PATH="$1"
SCORE=0
TOTAL=100

# Check file exists
if [ ! -f "$SKILL_PATH" ]; then
    echo "FAIL: Skill file not found"
    exit 1
fi

# Check frontmatter
if head -20 "$SKILL_PATH" | grep -q "^---"; then
    if grep -q "^name:" "$SKILL_PATH"; then
        SCORE=$((SCORE + 4))
    fi
    if grep -q "^description:" "$SKILL_PATH"; then
        SCORE=$((SCORE + 3))
    fi
    if grep -q "^version:" "$SKILL_PATH"; then
        SCORE=$((SCORE + 3))
    fi
fi

# Check name is kebab-case
NAME=$(grep "^name:" "$SKILL_PATH" | sed 's/name: *//')
if echo "$NAME" | grep -qE "^[a-z][a-z0-9-]*[a-z0-9]$"; then
    SCORE=$((SCORE + 5))
fi

# Check description has triggers
if grep -qi "trigger" "$SKILL_PATH"; then
    SCORE=$((SCORE + 10))
fi

# Check workflow section
if grep -qE "^## (Workflow|Steps|Process)" "$SKILL_PATH"; then
    SCORE=$((SCORE + 10))
fi

# Check code templates (min 2 code blocks)
CODE_BLOCKS=$(grep -c '```' "$SKILL_PATH")
if [ "$CODE_BLOCKS" -ge 4 ]; then  # 4 = 2 complete blocks (open + close)
    SCORE=$((SCORE + 15))
fi

# Check validation section
if grep -qE "^## (Validation|Verification|Checklist)" "$SKILL_PATH" || grep -q "\- \[ \]" "$SKILL_PATH"; then
    SCORE=$((SCORE + 10))
fi

# Check token efficiency
LINES=$(wc -l < "$SKILL_PATH")
SIZE=$(stat -f%z "$SKILL_PATH" 2>/dev/null || stat -c%s "$SKILL_PATH")
if [ "$LINES" -lt 2000 ] && [ "$SIZE" -lt 51200 ]; then
    SCORE=$((SCORE + 10))
fi

# Check has examples
if grep -qi "example" "$SKILL_PATH"; then
    SCORE=$((SCORE + 5))
fi

# Output result
echo "SKILL QUALITY SCORE: $SCORE/100"

if [ "$SCORE" -ge 70 ]; then
    echo "RESULT: APPROVED"
    exit 0
else
    echo "RESULT: REJECTED"
    exit 1
fi

AGENT Quality Criteria

Required Structure (30%)

---
name: agent-name             # REQUIRED
description: |               # REQUIRED
  What this agent does
tools: Read, Write, Edit     # REQUIRED: Minimal necessary tools
model: sonnet                # REQUIRED: haiku|sonnet|opus
---

# Agent Instructions         # REQUIRED

## When to Use              # REQUIRED
## Workflow                 # REQUIRED
## Success Criteria         # REQUIRED
## Failure Handling         # REQUIRED

Quality Checks

Model Selection Validation

TASK COMPLEXITY → MODEL MAPPING:

Simple Tasks (haiku):
- File search, pattern matching
- Simple code generation
- Documentation updates
- Format conversions

Medium Tasks (sonnet):
- Feature implementation
- Bug fixes
- Code review
- Test writing
- Most development work

Complex Tasks (opus):
- Architecture design
- Security analysis
- Multi-phase planning
- Complex debugging
- Research and analysis

Validation Script

#!/bin/bash
# validate-agent.sh <agent-path>

AGENT_PATH="$1"
SCORE=0

# Check frontmatter fields
for FIELD in "name:" "description:" "tools:" "model:"; do
    if grep -q "^$FIELD" "$AGENT_PATH"; then
        SCORE=$((SCORE + 3))
    fi
done

# Check model is valid
MODEL=$(grep "^model:" "$AGENT_PATH" | sed 's/model: *//')
if echo "$MODEL" | grep -qE "^(haiku|sonnet|opus)$"; then
    SCORE=$((SCORE + 15))
fi

# Check tools aren't overly permissive
if ! grep -q "tools:.*\*" "$AGENT_PATH"; then
    SCORE=$((SCORE + 15))
fi

# Check required sections
for SECTION in "When to Use" "Workflow" "Success" "Fail"; do
    if grep -qi "$SECTION" "$AGENT_PATH"; then
        SCORE=$((SCORE + 10))
    fi
done

# Check for vague phrases (deduct if found)
if grep -qiE "(as needed|if necessary|when appropriate|etc\.)" "$AGENT_PATH"; then
    SCORE=$((SCORE - 10))
fi

echo "AGENT QUALITY SCORE: $SCORE/100"
[ "$SCORE" -ge 70 ] && echo "RESULT: APPROVED" && exit 0
echo "RESULT: REJECTED" && exit 1

HOOK Quality Criteria

Required Structure

{
  "hooks": {
    "PreToolUse|PostToolUse|Stop": [
      {
        "matcher": "tool == \"ToolName\" && ...",  // REQUIRED: Valid matcher
        "hooks": [
          {
            "type": "command",
            "command": "#!/bin/bash\n..."         // REQUIRED: Valid bash
          }
        ],
        "description": "What this hook does"      // REQUIRED
      }
    ]
  }
}

Quality Checks

Matcher Syntax Validation

VALID MATCHERS:
- tool == "Bash"
- tool == "Edit" && tool_input.file_path matches ".*\\.ts$"
- tool == "Write" && tool_input.content contains "TODO"

INVALID MATCHERS:
- tool = "Bash"           # Wrong operator (= vs ==)
- Tool == "bash"          # Case sensitive
- tool == Bash            # Missing quotes

Validation Script

#!/bin/bash
# validate-hook.sh <hooks-json-path>

HOOKS_PATH="$1"
SCORE=0

# Check valid JSON
if jq empty "$HOOKS_PATH" 2>/dev/null; then
    SCORE=$((SCORE + 15))
else
    echo "FAIL: Invalid JSON"
    exit 1
fi

# Check each hook has required fields
HOOK_COUNT=$(jq '[.hooks.PreToolUse, .hooks.PostToolUse, .hooks.Stop] | flatten | length' "$HOOKS_PATH")

for i in $(seq 0 $((HOOK_COUNT - 1))); do
    # Check matcher exists
    MATCHER=$(jq -r ".hooks | to_entries | .[].value[$i].matcher // empty" "$HOOKS_PATH")
    if [ -n "$MATCHER" ]; then
        SCORE=$((SCORE + 5))
    fi

    # Check description exists
    DESC=$(jq -r ".hooks | to_entries | .[].value[$i].description // empty" "$HOOKS_PATH")
    if [ -n "$DESC" ]; then
        SCORE=$((SCORE + 3))
    fi
done

# Validate bash syntax in commands
COMMANDS=$(jq -r '.. | .command? // empty' "$HOOKS_PATH")
while IFS= read -r cmd; do
    if [ -n "$cmd" ]; then
        if echo "$cmd" | bash -n 2>/dev/null; then
            SCORE=$((SCORE + 10))
        fi
    fi
done <<< "$COMMANDS"

# Normalize score to 100
SCORE=$((SCORE > 100 ? 100 : SCORE))

echo "HOOK QUALITY SCORE: $SCORE/100"
[ "$SCORE" -ge 70 ] && echo "RESULT: APPROVED" && exit 0
echo "RESULT: REJECTED" && exit 1

Quality Gate Integration

Phase 6.5: COMPONENT VALIDATION (NEW)

After generating components (Phase 5) and before testing (Phase 6):

┌─────────────────────────────────────────────────────────────────────────────┐
│                     PHASE 6.5: COMPONENT VALIDATION                          │
│                                                                             │
│  FOR EACH generated component:                                              │
│  ┌─────────────────────────────────────────────────────────────────────────┐│
│  │  1. VALIDATE structure and syntax                                       ││
│  │  2. SCORE against quality criteria (0-100)                              ││
│  │  3. GRADE: A (90+), B (80-89), C (70-79), D (60-69), F (<60)           ││
│  │  4. DECISION:                                                           ││
│  │     - A/B/C: APPROVED → Proceed                                         ││
│  │     - D/F: REJECTED → Regenerate (max 3 attempts)                       ││
│  └─────────────────────────────────────────────────────────────────────────┘│
│                                                                             │
│  OUTPUT: component-validation-report.json                                   │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

Validation Report Format

{
  "phase": "6.5",
  "timestamp": "2024-01-15T10:30:00Z",
  "components_validated": 5,
  "results": [
    {
      "type": "skill",
      "name": "express-patterns",
      "path": ".claude/skills/express-patterns/SKILL.md",
      "score": 85,
      "grade": "B",
      "decision": "APPROVED",
      "checks": {
        "frontmatter_valid": true,
        "name_kebab_case": true,
        "description_has_triggers": true,
        "has_workflow": true,
        "has_code_templates": true,
        "code_syntax_valid": true,
        "has_validation": false,
        "not_duplicate": true,
        "token_efficient": true,
        "has_examples": true
      },
      "warnings": ["Missing validation section"],
      "suggestions": ["Add ## Validation section with checklist"]
    },
    {
      "type": "agent",
      "name": "api-builder",
      "path": ".claude/agents/api-builder.md",
      "score": 62,
      "grade": "D",
      "decision": "REJECTED",
      "checks": {
        "frontmatter_valid": true,
        "model_appropriate": false,
        "tools_minimal": false,
        "has_when_to_use": true,
        "has_workflow": false,
        "has_success_criteria": false,
        "has_failure_handling": false,
        "instructions_unambiguous": true
      },
      "errors": [
        "Model 'opus' too heavy for simple API generation",
        "Tools list too broad: includes all tools",
        "Missing workflow section",
        "Missing success criteria"
      ],
      "regeneration_hints": [
        "Use 'sonnet' model for API building",
        "Limit tools to: Read, Write, Edit, Bash",
        "Add step-by-step workflow",
        "Define what 'success' looks like"
      ]
    }
  ],
  "summary": {
    "approved": 4,
    "rejected": 1,
    "regeneration_required": true
  }
}

Regeneration Protocol

When a component is REJECTED:

ATTEMPT 1:
├── Analyze rejection reasons
├── Apply regeneration hints
├── Regenerate component
└── Re-validate

ATTEMPT 2:
├── If still rejected, simplify scope
├── Focus on core functionality only
├── Regenerate with minimal features
└── Re-validate

ATTEMPT 3:
├── If still rejected, use template
├── Copy nearest working component
├── Adapt template to requirements
└── Re-validate

ATTEMPT 4+:
├── STOP
├── Log failure to build report
├── Mark component as MANUAL_REQUIRED
└── Continue with other components

Quality Grading Scale

Quick Reference: What Makes Components Production-Ready

Skills

• Discoverable: Clear triggers in description
• Actionable: Step-by-step workflow
• Copy-Paste Ready: Working code templates
• Verifiable: Success criteria defined

Agents

• Right-Sized: Model matches complexity
• Minimal Tools: Only what's needed
• Clear Instructions: No ambiguity
• Handles Failure: Knows when to stop

Hooks

• Valid Syntax: JSON + Bash parse correctly
• Targeted: Precise matchers, no false positives
• Safe: Error handling, no destructive defaults
• Documented: Purpose is clear

Triggers

Use this skill when:

• Generating new skills/agents/hooks (Phase 5)
• Validating component quality (Phase 6.5)
• Reviewing existing components for improvement
• Debugging why a generated component doesn't work