Claude Code Plugins

Community-maintained marketplace

Feedback

claims-validator

@jmagly/ai-writing-guide
45
0

Validate documentation for unsupported claims, made-up metrics, and unverifiable statements. Use when relevant to the task.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name claims-validator
description Validate documentation for unsupported claims, made-up metrics, and unverifiable statements. Use when relevant to the task.

claims-validator

Validate documentation for unsupported claims, made-up metrics, and unverifiable statements.

Triggers

  • "check for unsupported claims"
  • "validate claims"
  • "review for BS"
  • "check metrics"
  • "verify claims in this document"
  • "find made-up stats"

Purpose

This skill identifies statements that make claims without evidence, including:

  • Performance metrics without benchmarks or data
  • Time/cost estimates without basis
  • Percentage claims without citation
  • Comparative statements without baselines
  • Features described as implemented that don't exist
  • Marketing superlatives presented as facts

Behavior

When triggered, this skill:

  1. Scans for metric claims:

    • Percentage improvements ("40% faster", "reduces by 60%")
    • Time estimates ("saves 2-3 hours", "in minutes not hours")
    • Cost projections ("$50-150/month", "ROI of 3x")
    • Performance numbers ("99x faster", "sub-millisecond")
  2. Identifies unsupported comparatives:

    • "faster than", "better than", "more efficient"
    • "best", "leading", "revolutionary", "game-changing"
    • "comprehensive", "complete", "full-featured"
  3. Checks for feature claims:

    • Commands or flags mentioned that don't exist in codebase
    • Features described in present tense that aren't implemented
    • Integration claims without actual integration code
  4. Validates citations:

    • Claims that reference data should have sources
    • Benchmarks should link to methodology
    • Statistics should be reproducible
  5. Generates report:

    • List each claim found
    • Classification (metric, comparative, feature, cost)
    • Recommendation (remove, add citation, verify, rephrase)

Claim Categories

Metrics Without Data

# Flagged
"Time Saved: 92-96% (9-15 hours → 45-60 minutes)"
"99x faster routing"
"45x cache speedup"

# Problem
No benchmark data, methodology, or reproducible test

# Fix
Remove claim, or add: "Based on [benchmark/test], measured [how]"

Cost Estimates Without Basis

# Flagged
"Budget $20-50/month for moderate use"
"Light usage: ~$10-20/month"
"Enterprise teams may see $100-500+/month"

# Problem
No actual usage data, varies wildly by use case

# Fix
Remove specific numbers, or link to pricing calculator/methodology

Time Estimates Without Data

# Flagged
"Deploy Full SDLC Framework (2 Minutes)"
"5 minutes, replaces 2-4 hours manual work"
"campaign setup from 2-3 weeks → 1 week"

# Problem
No measurement, varies by project complexity

# Fix
Remove time claims, describe what it does instead

Comparative Claims Without Baseline

# Flagged
"faster than manual processes"
"more efficient than traditional approaches"
"better than existing solutions"

# Problem
No specific comparison, no baseline defined

# Fix
Remove comparison, or specify exactly what's being compared

Feature Claims for Unimplemented Features

# Flagged
"aiwg -migrate-workspace  # Optional migration tool"
"Run 'config-validator --fix' to apply automated fixes"

# Problem
Command doesn't exist in codebase

# Fix
Remove until implemented, or mark as "Planned:"

Marketing Superlatives

# Flagged
"comprehensive", "revolutionary", "game-changing"
"best-in-class", "industry-leading", "cutting-edge"
"seamless", "effortless", "zero-friction"

# Problem
Subjective claims that can't be verified

# Fix
Replace with specific, factual descriptions

Validation Report Format

# Claims Validation Report

**Document**: README.md
**Date**: 2025-12-09
**Claims Found**: 12
**Issues**: 8

## Summary

| Category | Found | Unsupported | Action Needed |
|----------|-------|-------------|---------------|
| Metrics | 5 | 4 | Remove or cite |
| Time estimates | 3 | 3 | Remove |
| Cost estimates | 2 | 2 | Remove |
| Comparatives | 1 | 1 | Rephrase |
| Features | 1 | 1 | Remove (not implemented) |

## Issues

### 1. Unsupported Metric
**Line 204**: "Time Savings: 20-98% reduction across 5 core use cases"
**Problem**: No benchmark data or methodology
**Action**: Remove claim

### 2. Cost Estimate Without Data
**Line 1180**: "Light usage: ~$10-20/month"
**Problem**: No actual usage data to support estimate
**Action**: Remove specific numbers

### 3. Non-existent Feature
**Line 583**: "aiwg -migrate-workspace"
**Problem**: Command not implemented
**Action**: Remove until implemented

## Recommendations

1. **Remove all percentage-based time savings claims**
2. **Remove cost estimates** - direct users to measure their own usage
3. **Remove non-existent commands** - only document what exists
4. **Replace superlatives with specifics** - say what it does, not how good it is

Usage Examples

Validate a Document

User: "Check README.md for unsupported claims"

Output:
"Claims Validation: README.md

Found 8 unsupported claims:

Metrics (4):
- Line 204: '20-98% reduction' - no data
- Line 362: '56-63% time saved' - no data
- Line 588: '99x faster routing' - no benchmark
- Line 589: '45x cache speedup' - no benchmark

Time Estimates (2):
- Line 187: '5 minutes, replaces 2-4 hours' - unmeasured
- Line 366: '2 Minutes' in heading - varies by project

Non-existent Features (2):
- Line 583: 'aiwg -migrate-workspace' - not implemented
- Line 651: 'aiwg -rollback-workspace' - not implemented

Recommendation: Remove these claims. Describe what features
do, not how much time/money they save."

Quick Check

User: "Any BS in this section?"

Output:
"Found 2 issues:

1. 'Revolutionary approach' - marketing speak, be specific
2. '10x productivity gain' - no measurement

Suggest: Replace with factual descriptions of functionality."

Integration

This skill complements:

  • Voice Framework: Voice defines how to write, claims-validator checks what you claim
  • config-validator: Validates config files, claims-validator validates prose claims

What This Skill Does NOT Flag

  • Factual descriptions of features that exist
  • Documented benchmarks with methodology
  • Qualified statements ("may vary", "depending on", "in our testing")
  • User testimonials clearly attributed
  • Comparative claims with specific baselines cited

Output Location

  • Validation reports: .aiwg/reports/claims-validation.md