Imported from JayKim88/claude-ai-engineering (plugins/spec-validator/v3/skills/spec-validator/SKILL.md). Install upstream with npx skills add JayKim88/claude-ai-engineering --skill spec-validator. Copyright stays with the author.

Spec Validator Skill

Trigger Phrases

English:

• "validate spec implementation"
• "check if spec is implemented"
• "verify spec compliance"
• "validate against spec"
• "check spec checklist"
• "how complete is the implementation"
• "spec validation report"

Korean:

• "스펙 검증"
• "구현 확인"
• "체크리스트 검증"
• "명세 확인"
• "요구사항 검증"

When to Use

Use this skill when:

1. You have a spec document and want to verify implementation progress
2. You need to check if a .spec-checklist.yaml file is synchronized with the spec
3. You want to generate a scored validation report (100-point scale, 4 dimensions)
4. You need to track validation history and score improvement trends
5. You want actionable feedback on what's missing or incomplete
6. You're in a spec-driven development workflow and need periodic verification

Path Resolution Strategy

This skill uses a dynamic path detection system to avoid hardcoded absolute paths:

Step 0: Detect Skill Directory

# At the start of execution, determine where the skill files are located
# Use Bash tool to find the plugin installation directory
Bash(command="pwd", description="Get current working directory")
# Store result as current_directory

# The skill directory will be relative to the plugin root
# Typical structure: {plugin_root}/skills/spec-validator/
# Config path: {skill_directory}/config.yaml
# Templates path: {plugin_root}/templates/

Throughout Execution:

• Use user-provided absolute paths for spec and checklist files
• Use detected skill_directory for config.yaml
• Use detected plugin_root for templates
• Generate output paths relative to spec directory

Execution Algorithm

Step 1: Initialization and Context Gathering

Goal: Collect required inputs from user or context.

If spec path NOT provided in user request: Use AskUserQuestion tool to gather information:

AskUserQuestion(question="Please provide the path to your specification document (e.g., ./docs/spec.md):")

If checklist path NOT provided: Default to looking for .spec-checklist.yaml in the same directory as the spec file.

If validation mode NOT provided: Default to full mode.

Generate session ID: Use current timestamp (format: YYYYMMDD-HHMMSS)

Detect skill directory:

# Use Bash to detect where this skill is installed
Bash(command="pwd", description="Detect current working directory")
# If in plugin context, skill_directory will be determinable from context
# Otherwise, search for config.yaml in typical plugin paths

Outputs from this step:

• spec_path: Full absolute path to spec document
• checklist_path: Full absolute path to checklist (or default location)
• validation_mode: "full", "quick", or "requirements-only"
• session_id: Timestamp string
• skill_directory: Detected path to skill files
• config_path: {skill_directory}/config.yaml
• template_directory: {skill_directory}/../../templates

Branching Logic:

• If spec path is invalid → Use AskUserQuestion to get correct path
• If validation mode is invalid → Default to "full" and notify user
• Proceed to Step 2

Step 2: Load Configuration

Goal: Read configuration file to get scoring weights and parsing patterns.

Action: Use Read tool to load config:

Read(file_path=config_path)

Where config_path was determined in Step 1.

Parse the YAML content to extract:

• dimensions section (weights for each dimension)
• priority_weights (high=3, medium=2, low=1)
• grades thresholds (A=90-100, B=80-89, etc.)
• issue_severity mappings
• parsing.functional_requirements.patterns (regex patterns)
• validation_modes settings

If config file not found or malformed: Use these hardcoded defaults:

dimensions:
  implementation_completeness: 40
  implementation_quality: 25
  spec_adherence: 20
  progress_transparency: 15
priority_weights:
  high: 3
  medium: 2
  low: 1
  default: 2

Output to user: "Warning: Config not found. Using default settings."

Outputs from this step:

• config object with all settings

Proceed to Step 3

Step 3: Parse Specification Document

Goal: Extract FRs, NFRs, models, APIs, and edge cases from spec.

Action: Use Read tool:

Read(file_path=spec_path)

Parse the spec content using these strategies:

For Functional Requirements:

• Search for patterns like ## FR-1:, **FR-2**:, FR-3:
• For each match, extract:
  • ID (e.g., "FR-1")
  • Title (text after the colon)
  • Priority: Search nearby lines for "Priority: High", "(High)", "[HIGH]", etc.
    • If not found, default to "medium"
• Store in list: functional_requirements = [{"id": "FR-1", "title": "...", "priority": "high"}, ...]

For Non-Functional Requirements:

• Search for patterns like ## NFR-1:, **NFR-2**:, etc.
• Extract ID and title
• Store in list: non_functional_requirements = [{"id": "NFR-1", "title": "..."}, ...]

For Data Models:

• Find section headers: "## Technical Design", "## Data Models"
• Extract model names from patterns like ### User Model, #### Product
• Store in list: data_models = [{"name": "User"}, {"name": "Product"}, ...]

For API Endpoints:

• Search for patterns: GET /api/, POST /api/, etc.
• Also check: `GET /api/...` (in code blocks)
• Extract method and path
• Store in list: api_endpoints = [{"method": "GET", "path": "/api/users"}, ...]

For Edge Cases:

• Find section: "## Edge Cases", "### Edge Cases"
• Extract numbered or bulleted items
• Store in list: edge_cases = [{"id": "EC-1", "description": "..."}, ...]

Outputs from this step:

• spec_data object containing:
  • functional_requirements (list)
  • non_functional_requirements (list)
  • data_models (list)
  • api_endpoints (list)
  • edge_cases (list)

Branching Logic:

• If spec file is empty → Output error: "Spec document is empty. Cannot validate." → Exit
• If no FRs found → Log warning: "No functional requirements found. Check spec format." → Continue
• Proceed to Step 4

Step 4: Read and Validate Checklist

Goal: Load checklist and verify structure.

Action: Use Read tool:

Read(file_path=checklist_path)

If file not found: Use AskUserQuestion:

AskUserQuestion(question="Checklist not found at [checklist_path]. Would you like to:\n1) Generate template from spec\n2) Specify different path\n3) Exit\n\nEnter choice (1/2/3):")

• If choice = 1: Generate template (see template generation logic below), then Exit or Continue
• If choice = 2: Ask for new path, retry
• If choice = 3: Exit

Parse YAML content: Verify these sections exist:

• metadata
• functional_requirements
• non_functional_requirements

Validate each FR item has:

• id field
• status field (value must be: completed, in_progress, not_started, or blocked)
• test_status field (value must be: completed, in_progress, or not_started)

Check synchronization:

• Create set of all spec item IDs: spec_ids = {fr.id for fr in spec_data.functional_requirements} + {nfr.id for nfr in spec_data.non_functional_requirements}
• Create set of all checklist item IDs: checklist_ids = {item.id for item in checklist.functional_requirements} + {item.id for item in checklist.non_functional_requirements}
• Find orphaned: orphaned_items = checklist_ids - spec_ids
• Find missing: missing_items = spec_ids - checklist_ids

Outputs from this step:

• checklist object with parsed data
• sync_report:
  • orphaned_items (list)
  • missing_items (list)
  • sync_percentage (float)

Branching Logic:

• If YAML syntax invalid → Output error with line number → Exit
• If structure invalid → Output error with missing sections → Exit
• If out of sync → Flag for dimension 3 scoring → Continue
• Proceed to Step 5

Step 5: Calculate Dimension 1 - Implementation Completeness (40 points)

Goal: Calculate completion score with priority weighting.

FR Scoring (20 points max):

weighted_completed = 0
weighted_total = 0

for each fr in spec_data.functional_requirements:
    weight = config.priority_weights[fr.priority]  # high=3, medium=2, low=1
    weighted_total += weight

    checklist_fr = find_in_checklist(fr.id)
    if checklist_fr.status == "completed":
        weighted_completed += weight

fr_percentage = weighted_completed / weighted_total if weighted_total > 0 else 0
fr_score = fr_percentage * 20

NFR Scoring (10 points max):

nfr_completed = count(nfr in checklist where status == "completed")
nfr_total = len(spec_data.non_functional_requirements)
nfr_score = (nfr_completed / nfr_total) * 10 if nfr_total > 0 else 10

Model Scoring (5 points max):

models_completed = count(model in checklist where status == "completed")
models_total = len(spec_data.data_models)
model_score = (models_completed / models_total) * 5 if models_total > 0 else 5

API Scoring (5 points max):

apis_completed = count(api in checklist where status == "completed")
apis_total = len(spec_data.api_endpoints)
api_score = (apis_completed / apis_total) * 5 if apis_total > 0 else 5

Total:

dimension_1_score = fr_score + nfr_score + model_score + api_score

Outputs from this step:

• dimension_1_score (0-40)
• completeness_breakdown with per-category scores

Proceed to Step 6

Step 6: Calculate Dimension 2 - Implementation Quality (25 points)

Goal: Assess quality indicators.

Test Coverage (15 points max):

requirements_with_tests = count(items in checklist where test_status == "completed")
total_requirements = len(all checklist items with test_status field)
test_coverage_score = (requirements_with_tests / total_requirements) * 15 if total_requirements > 0 else 0

Edge Case Coverage (7 points max):

edge_cases_handled = count(ec in checklist where status == "completed")
edge_cases_total = len(spec_data.edge_cases)
edge_case_score = (edge_cases_handled / edge_cases_total) * 7 if edge_cases_total > 0 else 7

Implementation Notes (3 points max):

items_with_notes = count(items in checklist where implementation_notes is not empty)
total_items = len(all checklist items)
notes_score = (items_with_notes / total_items) * 3

Total:

dimension_2_score = test_coverage_score + edge_case_score + notes_score

Outputs from this step:

• dimension_2_score (0-25)
• quality_breakdown

Proceed to Step 7

Step 7: Calculate Dimension 3 - Spec Adherence (20 points)

Goal: Verify synchronization.

Synchronization Score (12 points max):

total_spec_items = len(spec_ids)
error_count = len(orphaned_items) + len(missing_items)
sync_accuracy = 1 - (error_count / total_spec_items) if total_spec_items > 0 else 1
sync_score = sync_accuracy * 12

Priority Accuracy (5 points max):

priority_matches = 0
total_frs = len(spec_data.functional_requirements)

for each fr in spec_data.functional_requirements:
    checklist_fr = find_in_checklist(fr.id)
    if checklist_fr and checklist_fr.priority == fr.priority:
        priority_matches += 1

priority_score = (priority_matches / total_frs) * 5 if total_frs > 0 else 5

Structure Compliance (3 points max):

required_sections = ["metadata", "functional_requirements", "non_functional_requirements"]
has_all_sections = all(section in checklist for section in required_sections)
structure_score = 3 if has_all_sections else 0

Total:

dimension_3_score = sync_score + priority_score + structure_score

Outputs from this step:

• dimension_3_score (0-20)
• adherence_breakdown

Proceed to Step 8

Step 8: Calculate Dimension 4 - Progress Transparency (15 points)

Goal: Evaluate tracking hygiene.

Status Hygiene (5 points max):

valid_statuses = ["completed", "in_progress", "not_started", "blocked"]
items_with_valid_status = count(items where status in valid_statuses)
total_items = len(all checklist items)
status_hygiene_score = (items_with_valid_status / total_items) * 5

Blocker Documentation (4 points max):

blocked_items = items where status == "blocked"
blocked_with_reasons = count(blocked_items where blockers field is not empty)
blocker_score = (blocked_with_reasons / len(blocked_items)) * 4 if len(blocked_items) > 0 else 4

Implementation Notes (3 points max):

items_with_meaningful_notes = count(items where len(implementation_notes) > 10)
notes_score = (items_with_meaningful_notes / total_items) * 3

Validation History (3 points max):

history_entries = len(checklist.validation_history) if hasattr(checklist, 'validation_history') else 0
if history_entries >= 2:
    history_score = 3
elif history_entries == 1:
    history_score = 1
else:
    history_score = 0

Total:

dimension_4_score = status_hygiene_score + blocker_score + notes_score + history_score

Outputs from this step:

• dimension_4_score (0-15)
• transparency_breakdown

Proceed to Step 9

Step 9: Aggregate Scores and Determine Grade

Goal: Calculate overall score and assign grade.

Score aggregation based on mode:

if validation_mode == "full":
    overall_score = dimension_1_score + dimension_2_score + dimension_3_score + dimension_4_score
elif validation_mode == "quick":
    raw_score = dimension_1_score + dimension_3_score  # max 60
    overall_score = (raw_score / 60) * 100
elif validation_mode == "requirements-only":
    raw_score = dimension_1_score  # max 40
    overall_score = (raw_score / 40) * 100

Grade assignment:

if overall_score >= 90:
    grade = "A"
    label = "Ready to ship"
elif overall_score >= 80:
    grade = "B"
    label = "Nearly complete"
elif overall_score >= 70:
    grade = "C"
    label = "Good progress"
elif overall_score >= 60:
    grade = "D"
    label = "Needs work"
else:
    grade = "F"
    label = "Significant gaps"

Outputs from this step:

• overall_score (0-100)
• grade (A-F)
• grade_label (string)

Proceed to Step 10

Step 10: Generate Issues List

Goal: Categorize all findings.

Critical Issues (from config.issue_severity.critical):

• Missing high-priority FRs (where status != "completed")
• Orphaned checklist items (in checklist but not in spec)
• Invalid YAML structure
• Synchronization errors (>10% items out of sync)

Important Issues (from config.issue_severity.important):

• Missing medium-priority FRs
• Completed items without tests (status="completed" but test_status!="completed")
• Missing NFRs
• Priority mismatches between spec and checklist
• Blocked items without blocker explanations

Minor Issues (from config.issue_severity.minor):

• Missing low-priority FRs
• Incomplete implementation notes
• Missing validation history

For each issue, create:

{
  "title": "FR-X not completed",
  "category": "Functional Requirement",
  "item_id": "FR-X",
  "description": "High-priority requirement not implemented",
  "impact": "Blocks core functionality",
  "fix_action": "Implement FR-X in [relevant file]",
  "file_location": checklist_path
}

Outputs from this step:

• critical_issues (list)
• important_issues (list)
• minor_issues (list)
• next_steps (list of actions)

Proceed to Step 11

Step 11: Update Validation History

Goal: Append validation entry to checklist.

Action: Use Read tool to get current checklist:

Read(file_path=checklist_path)

Create history entry:

new_entry = {
  "date": current_timestamp_ISO8601,
  "score": overall_score,
  "grade": grade,
  "dimensions": {
    "completeness": dimension_1_score,
    "quality": dimension_2_score,
    "adherence": dimension_3_score,
    "transparency": dimension_4_score
  },
  "mode": validation_mode,
  "validator_version": "1.0.0"
}

Append to checklist YAML: If validation_history section exists, append to it. If not, create new section.

Calculate delta if previous history exists:

if len(validation_history) > 1:
    score_delta = overall_score - validation_history[-2].score
    if score_delta > 0:
        trend = "↗ Improving"
    elif score_delta < 0:
        trend = "↘ Declining"
    else:
        trend = "→ Stable"

Action: Use Write tool to save updated checklist:

Write(file_path=checklist_path, content=updated_checklist_yaml)

Outputs from this step:

• Updated checklist file
• score_delta (if applicable)
• trend (if applicable)

Branching Logic:

• If write fails → Log error: "Could not update validation history" → Continue to Step 12

Proceed to Step 12

Step 12: Generate Validation Report

Goal: Create comprehensive Markdown report.

Detect template path:

template_path = template_directory + "/validation-report-template.md"

Action: Use Read tool to load template:

Read(file_path=template_path)

Template Substitution Algorithm:

The template uses [PLACEHOLDER] format. To populate:

Step 12.1: Read Template

template_content = Read(file_path=template_path)

Step 12.2: Prepare All Replacement Values Create a dictionary mapping placeholders to actual values:

replacements = {
  "[TIMESTAMP]": current_timestamp,
  "[VALIDATION_MODE]": validation_mode,
  "[SPEC_PATH]": spec_path,
  "[CHECKLIST_PATH]": checklist_path,
  "[OVERALL_SCORE]": str(overall_score),
  "[GRADE]": grade,
  "[GRADE_DESCRIPTION]": grade_label,
  "[COMPLETENESS_SCORE]": str(dimension_1_score),
  "[COMPLETENESS_MAX]": "40",
  "[COMPLETENESS_PERCENTAGE]": str(int(dimension_1_score / 40 * 100)),
  # ... continue for all placeholders
}

Step 12.3: Handle Table Row Placeholders For placeholders ending in _TABLE_ROWS, generate markdown table rows:

# Example: Functional Requirements Table
fr_rows = []
for fr in functional_requirements:
    row = f"| {fr.id} | {fr.title} | {fr.priority} | {fr.status} | {fr.test_status} | {fr.completed_date or '-'} | {fr.implementation_notes or '-'} |"
    fr_rows.append(row)

replacements["[FUNCTIONAL_REQUIREMENTS_TABLE_ROWS]"] = "\n".join(fr_rows)

Step 12.4: Handle Conditional Sections For sections marked with [SECTION_NAME - If condition:], check condition and include/exclude:

# Example: Previous Score Section
if has_previous_validation:
    previous_section = f"**Previous Score**: {previous_score}/100 ({score_delta})\n**Trend**: {trend_indicator} {trend_label}"
    # Keep the section
else:
    # Remove the entire conditional section from template
    # Find the section between markers and delete it

Step 12.5: Perform String Replacement

report_content = template_content

# Replace all simple placeholders
for placeholder, value in replacements.items():
    report_content = report_content.replace(placeholder, value)

# Remove any remaining conditional markers
# (If a conditional section was kept, remove its markers)

Step 12.6: Validate Report Content Verify that:

• No [PLACEHOLDER] strings remain (except in usage instructions section)
• All tables have consistent column counts
• All scores are numeric and in valid ranges

Generate report filename:

report_filename = f"validation-report-{session_id}.md"
report_path = spec_directory + "/" + report_filename

Where spec_directory is the directory containing the spec file.

Action: Use Write tool to save report:

Write(file_path=report_path, content=report_content)

Output to user:

=== Spec Validation Complete ===

Overall Score: [overall_score]/100 (Grade [grade])
[grade_label]

Critical Issues: [critical_count]
Important Issues: [important_count]
Minor Issues: [minor_count]

Full report saved to: [report_path]
Validation history updated in checklist.

Branching Logic:

• If template not found → Use inline template structure → Continue
• If write fails → Output report to console instead → Continue

End of execution

Error Handling

Quick Reference

Command Usage

# Full validation (default)
validate spec implementation

# Quick validation (Completeness + Adherence only)
validate spec implementation in quick mode

# Requirements-only (FR/NFR progress only)
validate requirements only

# Specify paths explicitly
validate spec at /path/to/spec.md with checklist /path/to/.spec-checklist.yaml

# Korean
스펙 검증

Validation Modes

Scoring Dimensions

Grade Scale

• 90-100 (A): Ready to ship. High confidence in completeness.
• 80-89 (B): Nearly complete. Minor gaps, generally ready.
• 70-79 (C): Good progress. Significant work done, but notable gaps.
• 60-69 (D): Needs work. More incomplete than complete.
• 0-59 (F): Significant gaps. Major requirements missing.

Typical Workflow

1. Create spec with spec-interview plugin
2. Generate checklist: generate checklist from spec.md
3. Implement features, update checklist after each completion
4. Validate periodically: Every sprint or before releases
5. Fix issues identified in validation report
6. Re-validate until score >= 90
7. Ship with confidence

File Locations

• Spec document: Anywhere (user provides path)
• Checklist: Same directory as spec, named .spec-checklist.yaml
• Config: Detected at {skill_directory}/config.yaml
• Report: {spec_directory}/validation-report-{timestamp}.md
• Templates: Detected at {plugin_root}/templates/