MantaBase T3 Hardware Audit System

Task Objectives

This Skill is used to: automatically crawl information through hardware product links, after Brand Blinding (brand-removal) processing, three independent Auditors conduct special scoring based on authoritative reference books, optimize through Peer Review mutual examination, and finally achieve objective Tool/Toy/Trash classification
Capabilities include: web information crawling, Brand Blinding information filtering, three Auditor special scoring, Peer Review mutual examination, Final Judge comprehensive judgment
Trigger conditions: user provides hardware product link and requests T3 audit or classification scoring
Core concepts: transparency first, objective evaluation, parallel audit, expert review, continuous optimization

Preparations

No additional preparation required, the system will automatically extract and process web content

Operation Steps

Standard Process

1. Obtain User Input

Guide users to provide hardware product links (supports e-commerce platforms, official websites, review websites, etc.)
Optional: Obtain specific dimensions or audit focus that users care about

2. Obtain Product Information (Recommended Agent Method)

Mandatory Read references/mandatory-page-list.md, execute fixed crawling lists according to site types (Shopify / Kickstarter) to ensure consistent data input.

Method A: Direct Crawling by Agent (Recommended)

Read references/web-fetch-guide.md to master the complete extraction strategy
After identifying the site type, fetch sequentially according to the mandatory-page-list, and write to
```
01-level0-source-urls.md
```
,
```
02-level0-extracts.md
```
Use the web_fetch tool to directly access the product page
Advantages:
- ✅ High data integrity (intelligently extract key information)
- ✅ Adapt to different website structures
- ✅ Handle complex pages (dynamic loading, multi-tab pages)
- ✅ Access multiple related pages simultaneously (product page, review page, specification page)
- ✅ Agent can actively identify missing information and supplement it

Execution Points:

⚠️ Important: A single page may have incomplete information, the Agent should:

1. Multi-page access:
   - Product detail page → basic information, price
   - Product specification page → technical parameters
   - User review page → user feedback
   - Review page → professional evaluation

2. Price information extraction:
   - Find schema.org structured data
   - Try the shopping cart page
   - Search for "product name + price"
   - If missing: clearly mark in the report

3. Dynamic content processing:
   - Use pagination crawling (offset parameter)
   - Try different URL variants
   - Find mobile version pages

4. Data integrity verification:
   - Basic information: must be complete
   - Technical specifications: at least 50%
   - User feedback: at least one item

Method B: Python Script Crawling (Auxiliary/Batch Scenarios)

Applicable scenarios: batch processing of multiple products, offline analysis

Call the script to extract product page content:

bash

python scripts/crawl_product_info.py --url <product_url> --pretty

Parameter description:
- ```
--url
```
  : Product page URL (required)
- ```
--pretty
```
  : Beautify JSON output (optional)
- ```
--output
```
  : Output file path (optional)
Note: The script method may have incomplete data, it is recommended to use the Agent method first
Script returns: structured information such as product name, description, features, price, target users, complete objective data, etc.

3. Brand Blinding Information Filtering (Executed by Agent)

Read references/defluff-guide.md to master Brand Blinding principles
The Agent performs brand-removal processing on product information:
- Remove brand names and trademark references
- Remove marketing words and exaggerated descriptions
- Identify and remove emotional adjectives
- Extract pure functional information and objective parameters
- Maintain the integrity and accuracy of facts
Output: objective product information after Brand-Blinding

4. Three Auditor Special Scoring (Executed in Parallel by Agents)

Each Auditor evaluates independently simultaneously, performing their respective duties
Each Auditor reads their own independent guide, and does not view other Auditors' guides or reports

🟢 Tool Auditor

Read references/tool-auditor.md to obtain the complete guide
Mandatory Compliance references/auditor-templates.md strictly follow the scoring table template
Evaluation angles: solving actual pain points, practicality, reliability
Authoritative reference: Tony Fadell's Build special checklist
Special inspection:
- Pain point identification and resolution (30 points)
- Attention to detail and consistency (25 points)
- Conciseness and efficiency (25 points)
- Engineering reliability (20 points)
Output: special scoring report (total 100 points), including scores, reasons, and evidence for each inspection item
Litmus Test: If it breaks tomorrow, will the user's workflow be stagnated?

🟡 Toy Auditor

Read references/toy-auditor.md to obtain the complete guide
Mandatory Compliance references/auditor-templates.md strictly follow the scoring table template
Evaluation angles: emotional value, pleasant experience, aesthetic design
Authoritative reference: Don Norman's The Design of Everyday Things special checklist
Special inspection:
- Sensory pleasure (30 points)
- Surprise and discovery (25 points)
- Emotional connection (25 points)
- Explorability (20 points)
Output: special scoring report (total 100 points), including scores, reasons, and evidence for each inspection item
Litmus Test: Will users put it in a prominent position to "show off" or collect it?

🔴 Trash Auditor

Read references/trash-auditor.md to obtain the complete guide
Mandatory Compliance references/auditor-templates.md strictly follow the scoring table template
Mandatory Read references/trash-red-flags.md apply the "Sharp Eyes" high-sensitivity trigger list
Evaluation angles: logical defects, marketing fraud, design violations
Authoritative reference: Dieter Rams' ten design principles violation degree checklist
Special inspection:
- Principle violation (30 points)
- Problem creation (25 points)
- Value deficiency (25 points)
- Replaceability (20 points)
Output: special scoring report (total 0-100 points, positive scoring), including scores, reasons, and evidence for each inspection item
Litmus Test: If this product disappears tomorrow, will the world be better or worse?

Key Principles:

Information Isolation: Each Auditor can only access the objective product information after Brand-Blinding, and is strictly prohibited from viewing the original product information
Special Scoring: Must score based on the special checklist of authoritative reference books
Evidence-Driven: Each score must be based on specific evidence, avoiding subjective judgment
Independent Evaluation: The three Auditors evaluate simultaneously, without interfering with or communicating with each other
Objective Reporting: Each Auditor only reports objective findings, and records truthfully even if the evidence supports other classifications
Honest Recording: Even if the evidence supports other classifications, it must be recorded truthfully

5. Peer Review Mutual Examination (Executed in Parallel by Agents)

Read references/peer-review-guide.md to master the Peer Review process
The three Auditors mutually examine the reports of the other two Auditors:

Cross-Examination:

Tool Auditor examines: Toy report + Trash report
Toy Auditor examines: Tool report + Trash report
Trash Auditor examines: Tool report + Toy report

Examination Content:

Is the scoring reasonable?
Is the evidence sufficient?
Is there any overlooked cross-classification evidence?
Is there any bias or omission?

Examination Output:

Examination opinions (including specific suggestions and reasons)
Scoring adjustment suggestions
Cross-classification evidence supplement

Key Principles:

Based on facts and evidence, not targeting individuals
Provide specific improvement suggestions
Examiners and examinees maintain an open mindset
Record all examination opinions and responses

6. Auditor Optimize Report (Executed by Agent)

Each Auditor optimizes their own report based on the received examination opinions:
- Analyze the rationality of the examination opinions
- Accept reasonable suggestions and modify the report
- Provide refutation reasons for unreasonable opinions
- Record all modifications and reasons

Optimization Output:

Original report
Optimized report
Modification records (accepted/rejected examination opinions and reasons)

7. Final Judge Comprehensive Judgment (Executed by Agent)

Read references/t3-classification.md to obtain the classification judgment guide
Read the optimized final reports of the three Auditors
Comprehensive analysis of the conclusions and evidence of the three reports
Apply T3 classification rules:
- Calculate comprehensive score: Composite = max(Tool score, Toy score) - Trash score
- Determine the dominant classification (meet more than 2 conditions)
- Determine the secondary classification (if any)
- Apply Litmus Test consistency analysis
Output: final classification, confidence level, judgment reasons, improvement suggestions

8. Generate Audit Report

Output File: Unifiedly named

99-audit-report.md

(aliases such as

99-report-summary.md

are prohibited). Report directory:

tmp/reports/t3-{YYYY-MM-DD}-{case-id}/

Report Structure (see references/report-schema.md):

YAML Metadata Block (wrapped with
```
---
```
): parsed by the website leaderboard, including
```
case_id
```
,
```
source_url
```
,
```
scores
```
,
```
chart_data
```
,
```
litmus_tests
```
,
```
classification
```
, etc.
Text Analysis Content: aggregated from the
```
extract_for_report
```
of the three Auditors, fixed chapters: product overview, Tool key points, Toy key points, Trash key points, Final Judge judgment reasons, improvement suggestions

Must Include:

Basic product information (comparison between original information and Brand-Blinded information)
Three Auditor special scoring table (format specified in auditor-templates.md)
Peer Review summary
Final Judge judgment result

Optional Branches

When product information is insufficient: prompt the user to supplement the description or provide more reference materials
When scores are close to the boundary: analyze the dominant factors and explain the classification judgment basis
When Auditors have divergent opinions: Final Judge must explain the trade-off logic in detail
When there is controversy: provide multi-perspective analysis, list the reasons and evidence for different classifications

Resource Index

Auxiliary Scripts (batch processing/offline analysis scenarios):
- scripts/crawl_product_info.py (Purpose: web content crawling and product information extraction, parameters:
```
--url <product_url> [--pretty] [--output <file>]
```
  )
- scripts/synthesize_results.py (Purpose: format the three Auditor results and calculate the final classification, parameters:
```
--input <auditor_reports.json> [--pretty] [--output <file>]
```
  )
Domain References:
- references/mandatory-page-list.md (When to read: step 2 Obtain Product Information, mandatory read; fixed crawling list)
- references/report-schema.md (When to read: step 8 Generate 99-audit-report.md, mandatory compliance; YAML structure definition)
- references/file-naming-convention.md (When to read: when creating the report directory, mandatory compliance; file naming convention)
- references/isolation-manifest-template.md (When to read: reference when creating 00-isolation-manifest.md)
- references/web-fetch-guide.md (When to read: when the Agent uses the web_fetch tool to extract product information, mandatory read)
- references/tool-auditor.md (When to read: when Tool Auditor performs special scoring)
- references/toy-auditor.md (When to read: when Toy Auditor performs special scoring)
- references/trash-auditor.md (When to read: when Trash Auditor performs special scoring)
- references/trash-red-flags.md (When to read: when Trash Auditor applies the Sharp Eyes trigger list, mandatory read)
- references/auditor-templates.md (When to read: when the three Auditors perform special scoring, mandatory compliance; ensure reproducible table output)
- references/t3-classification.md (When to read: when Final Judge conducts comprehensive judgment)
- references/scoring-checklists.md (When to read: when quickly checking the checklist index)
- references/defluff-guide.md (When to read: when performing Brand Blinding processing)
- references/peer-review-guide.md (When to read: when performing Peer Review)
- references/objective-data-standard.md (When to read: all stages, ensure objective data integrity)
- references/design-theories.md (When to read: when conducting design theory analysis)

Notes

🚨 Information Isolation (Most Important):
- Must ensure that the three Auditors can only access the objective product information after Brand-Blinding
- Strictly prohibit Auditors from viewing original product information, brand names, and marketing words
- When executing Auditor tasks, the Agent must clearly state that the evaluation is only based on Brand-Blinded information
- The audit report must include the "information source description" field
📊 Objective Data Integrity (Core):
- Must collect complete objective data (technical specifications, performance, reliability, market, sustainability, cost)
- All objective data must be retained during the Brand Blinding stage, and cannot be removed or modified
- Each Auditor's score must be based on verifiable objective data, avoiding subjective speculation
- Refer to references/objective-data-standard.md to understand the objective data standard
- The audit report must include objective data summary and data integrity score
- If objective data is insufficient, must mark the impact of data missing on the evaluation
Importance of Brand Blinding:
- Must perform Brand Blinding first to ensure the evaluation is based on objective information rather than brand influence
- Brand Blinding must retain all objective data, only removing brand information and marketing words
- Distinguishing between "objective data" and "marketing words" is the core task of Brand Blinding
Necessity of Special Scoring:
- Must score based on the special checklist of authoritative reference books
- Each score must be traced back to specific inspection items and evidence
- Ensure the consistency and repeatability of scoring
Independence of Three Auditors:
- The three Auditors must evaluate simultaneously and independently, without interfering with or communicating with each other
- Each Auditor can only see their own evaluation task and Brand-Blinded information
- Prohibit Auditors from exchanging information or referring to other Auditors' conclusions
Objectivity of Peer Review:
- Examination is based on facts and evidence, not targeting individuals
- Provide specific improvement suggestions
- Examiners and examinees maintain an open mindset
- Record all examination opinions and responses
Objectivity Principle: Auditors report objective findings, and must report truthfully even if the evidence supports other classifications
Transparency: All evaluation processes, reasons, and evidence must be clearly presented in the report
Agent-Led: Brand Blinding, three Auditor special scoring, Peer Review, and Final Judge are all executed by Agents, and scripts are only responsible for crawling and formatting

Usage Examples

Example 1: Complete T3 Audit Process

Function Description: Complete Brand Blinding + special scoring + Peer Review + Final Judge process
Execution Method:
1. Agent crawls product information
2. Agent executes Brand Blinding
3. Agent simultaneously starts three Auditor special scoring
4. Agent executes Peer Review mutual examination
5. Agent optimizes the report
6. Agent executes Final Judge judgment
7. Generate complete audit report
Key Parameters: product URL

Output Example:

Product: Smart Watch X
Brand-Blinded: Wearable health monitoring device with heart rate, blood oxygen, sleep tracking functions

🟢 Tool Auditor Special Score: 82/100
  - Pain point identification and resolution: 28/30 ✓
    * Identified clear pain points (9/10)
    * Solution directly addresses pain points (9/10)
    * Better than existing solutions (10/10)
  - Attention to detail and consistency: 20/25 ✓
    * Interaction is carefully designed (8/10)
    * Design language is consistent (7/10)
    * Hides complexity (5/5)
  - Conciseness and efficiency: 20/25 ✓
    * Follows less is more (8/10)
    * Operation is efficient (7/10)
    * Unobtrusive (5/5)
  - Engineering reliability: 14/20 ✓
    * Stable and reliable (7/10)
    * Quality control (7/10)

🟡 Toy Auditor Special Score: 65/100
  - Visibility: 12/15 ✓
    * Status is clearly visible (8/10)
    * Operation is obvious (4/5)
  - Feedback: 11/15 ✓
    * Immediate feedback (7/10)
    * Feedback information is accurate (4/5)
  - Constraints: 12/15 ✓
    * Constraint mechanism is perfect (7/10)
    * Constraints are natural and reasonable (5/5)
  - Mapping: 11/15 ✓
    * Natural mapping (7/10)
    * Conforms to mental model (4/5)
  - Consistency: 7/10 ✓
    * Internally consistent (4/5)
    * Conforms to standards (3/5)
  - Emotional design: 12/30 ✓
    * Visceral level (6/10)
    * Behavioral level (3/10)
    * Reflective level (3/10)

🔴 Trash Auditor Special Score: 25/100
  - Principle 1 (Innovation): 2/10 - Minor innovation
  - Principle 2 (Usefulness): 8/10 - Practical functions
  - Principle 3 (Beauty): 6/10 - Average appearance
  - Principle 4 (Understandability): 7/10 - Basically easy to understand
  - Principle 5 (Unobtrusiveness): 8/10 - Basically unobtrusive
  - Principle 6 (Honesty): 3/10 - Slightly exaggerated
  - Principle 7 (Durability): 5/10 - May become obsolete
  - Principle 8 (Detail): 6/10 - Average details
  - Principle 9 (Environmental friendliness): 4/10 - Average environmental protection
  - Principle 10 (As little as possible): 7/10 - Relatively concise

Peer Review Results:
  - Tool Auditor suggests reducing Toy score (-15 points) ✓ Accepted
  - Toy Auditor suggests increasing Tool score (+5 points) ✗ Rejected
  - Trash Auditor suggests reducing Toy score (-10 points) ✓ Accepted

Final Judge: Tool (Confidence: 78%)
Reason: Tool special score (82) is significantly higher than Toy special score (65), Trash score (25) is low, the main value of the product lies in solving health monitoring pain points

Example 2: Quick T3 Classification

Function Description: Simplified process, focusing on Brand Blinding and key special scoring
Execution Method: Agent simplifies the evaluation, focusing on core inspection items
Key Parameters: product URL, evaluation focus (optional)
Guidelines: Quickly identify the core value proposition of the product, apply Litmus Test and special checklists

Example 3: Competitor Comparison Audit

Function Description: Conduct parallel T3 audit comparison on multiple similar products
Execution Method: batch crawling + parallel three Auditor special scoring + Peer Review + comparative analysis
Output Format: comparison table + detailed audit report for each product

Audit Process Visualization

User Input URL
    ↓
Crawl Product Information (Original Information)
    ↓
Brand Blinding (Agent removes brand influence)
    ├─ Original Information 🔒 (Sealed, not accessible to Auditors)
    └─ Brand-Blinded Information ✅ (Only information source accessible to Auditors)
    ↓
┌─────────────────────────────────────────┐
│ Three Auditor Special Scoring (Parallel Execution by Agents) │
│ 🔒 Information Isolation: Each Auditor can only access │
│    Brand-Blinded objective product information           │
│ 📚 Authoritative Reference: Special Scoring Checklist │
├───────────────┬─────────────┬────────────┤
│ 🟢 Tool       │ 🟡 Toy      │ 🔴 Trash   │
│ Auditor       │ Auditor     │ Auditor    │
│ 🔒 Information Isolation   │ 🔒 Information Isolation │ 🔒 Information Isolation│
│ 📚 Build Checklist  │ 📚 The Design of Everyday Things │ 📚 Rams' Principles│
└───────┬───────┴─────┬───────┴──────┬────┘
        │             │              │
   Special Scoring      Special Scoring       Special Scoring
   (Initial Version)   (Initial Version)     (Initial Version)
        │             │              │
   Independent Report      Independent Report       Independent Report
        │             │              │
        └─────────────┼──────────────┘
                      ↓
┌─────────────────────────────────────┐
│  Peer Review Mutual Examination (Parallel Execution by Agents) │
│  🔒 Information Isolation: Only can view the other party's De- │
│     Branded report, cannot view original information      │
├─────────────────────────────────────┤
│ Tool examines Toy+Trash │ Toy examines Tool+Trash│
│  │Examination Opinions         │ Examination Opinions        │
│  └─────────────────┴──────────────┐ │
│  Trash examines Tool+Toy │ Examination Opinions      │ │
└───────┬─────────────┴──────────────┘
        │
   Examination Opinion Summary
        ↓
┌─────────────────────────────────────┐
│  Auditor Optimize Report (Agent)           │
├─────────────────────────────────────┤
│ Tool Optimization │ Toy Optimization │ Trash Optimization      │
└───────┬────┴────┬───┴─────┬─────────┘
        │          │          │
   Optimized Report   Optimized Report   Optimized Report
        │          │          │
        └──────────┼──────────┘
                   ↓
         Final Judge Comprehensive Judgment
                   ↓
               Generate Audit Report
               (Includes complete audit chain)

🔒 Information Isolation Guarantee

To ensure the objectivity of the three Auditors, the system must strictly follow the following information isolation rules:

Information Levels

Original Information (Level 0): includes brand names, marketing words, emotional descriptions
- Status: 🔒 Sealed
- Access Permissions: Only Brand Blinding module and Final Judge (for comparison)
- Prohibition: Three Auditors are strictly prohibited from accessing
Brand-Blinded Information (Level 1): pure functional and objective parameters
- Status: ✅ Public
- Access Permissions: Three Auditors and Final Judge
- Content: technical specifications, function lists, usage scenarios (brand-removed)

Agent Execution Checklist

When executing Auditor tasks, the Agent must confirm:

Has clearly stated that only Brand-Blinded information is used
Has scored based on the special checklist of authoritative reference books
Has not quoted any brand names
Has not referred to any marketing words
The evaluation report includes the statement "Information Source: Brand-Blinded product information"
The evaluation report includes the statement "Special Scoring Basis: XXX Checklist"

Audit Report Verification

Each Auditor's report must include:

json

{
  "auditor": "Tool",
  "total_score": 82,
  "checklist_score": 82,
  "information_source": "Brand-Blinded product information (original brand information has been isolated)",
  "scoring_basis": "Tony Fadell's *Build* Special Checklist",
  "information_isolation_confirmed": true,
  "checklist_compliance_confirmed": true
}

t3-hardware-scoring-cn

NPX Install

Tags

SKILL.md Content (Chinese)