An automated document review and grading agent designed for comprehensive, multi-dimensional academic thesis verification and cross-document validation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-grading
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install auto-grading using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Auto-Grading and Multi-Document Review Skill?
The Auto-Grading skill is a powerful automation capability designed for AI coding agents to streamline the grading and compliance review of graduation designs, thesis papers, and multi-document archives. It orchestrates a sophisticated document parsing workflow that handles various formats including doc, docx, pdf, and pptx, subjecting them to a rigorous multi-dimensional evaluation covering content integrity, structural formatting, logical consistency, and reference standards.
Built with a production-grade safety mindset, this Openclaw Skills asset implements a rigorous multi-tiered Harness Engineering framework that enforces a 'prefer omission over false accusation' policy. It programmatically guarantees that any identified discrepancy or grading penalty is accompanied by factual, verifiable evidence cross-referenced directly from the source text, entirely mitigating the risk of LLM hallucinations during grading processes.
Auto-Grading and Multi-Document Review Skill Use Cases
- Batch processing and auditing large volumes of student graduation designs or academic thesis archives.
- Executing cross-document consistency checks on related project documents (e.g., matching student IDs, supervisor names, and project titles across proposals, final reports, and slides).
- Automating GB/T 7714 bibliography format verification and internal citation mapping.
- Reviewing presentation materials (pptx) for structural flow, information density, visual balance, and key argument expression.
How Auto-Grading and Multi-Document Review Skill Works
- Batch Extraction: The agent executes the doc-browser pattern to batch-read files from the target directory or zip archive using specialized libraries based on file type (.docx, .doc, .pdf, .pptx). It saves clean text representations to a temporary directory.
- Single-Document Evaluation: The agent evaluates each core document independently against academic templates (
references/review_template.md), assigning metric-specific scores for integrity, format, logic, and reasoning. - Cross-Document Cross-Checking: The agent performs deep relational validation across files to verify administrative data continuity, chronological correctness of milestones, and reference synchronization.
- Defense Harness Application: The raw issue array is passed through a three-layer validation script (
harness.py) that checks format compliance, conducts an automated text grep to verify empirical evidence, and handles document version conflicts. - Structured Report Generation: The agent combines the verified issue logs and normalized score tables into a structured Markdown summary containing detailed assessment reports and final grading metrics.
Auto-Grading and Multi-Document Review Skill Setup
Install the required core python dependencies into your active environment:
pip install PyPDF2 python-docx python-pptx olefile
Ensure your system includes underlying binary extractors if parsing legacy .doc formats (macOS provides textutil natively; antiword can optionally be configured for deep fallback). Place your grading standards template inside references/review_template.md before execution.
Auto-Grading and Multi-Document Review Skill Data Schema & Taxonomy
The skill organizes its operation around standard file paths and structured JSON contracts:
- Intermediary Text Cache: Raw extractions are mapped to
/tmp/auto_grading/[filename].txt. - Review Template: Custom evaluation parameters are drawn from
references/review_template.md. - Output Manifest (
issues.json): Contains structured metadata with the following schema:
| Field Name | Type | Description |
|---|---|---|
student |
String | Extracted student name. |
single_doc_scores |
Object | Dimensional scores for evaluated materials. |
cross_check |
Object | Metric scores representing system consistency. |
issues |
Array | Objects containing file, issue, evidence, and confidence strings. |
Auto-Grading and Multi-Document Review Skill Advanced Features
- Harness Engineering Safety Guard: Built-in Python execution layer that programmatically drops or downgrades issues if the specific underlying reference string cannot be natively grepped from the source text.
- Smart Version Resolution: L3 version arbitration rules automatically identify and skip placeholder half-finished templates when duplicate
.docand.docxfile paths exist concurrently. - Field Residual Remnant Detection: Scans Microsoft Word XML hierarchies for field artifacts such as TOC, HYPERLINK, and PAGEREF residuals to catch broken formatting before finalizing reports.
SKILL.md
Loading
Related Openclaw Skills
Dochost Publish
A powerful skill that allows AI agents to instantly publish Markdown or HTML content into live, shareable dochost.io web pages.
Jianwei Strategic Double-Engine Insight System
Jianwei is a sophisticated strategic analysis skill powered by a double-engine architecture that generates, validates, and audits business hypotheses through rigorous frameworks.
PowerMatrix GEO Growth Orchestrator
A unified Generative Engine Optimization (GEO) workflow that audits brand visibility across AI models and structures platform-ready content.
Webhook Relay
A powerful utility that moves HTTP traffic across network boundaries, enabling secure webhook forwarding, local server tunneling, and payload debugging without opening firewall ports.
Super Mario Innovation & Brand Strategy Skill
An AI-powered business and design coach based on the story of Nintendo's rise to global dominance, helping you build resilient brands and simple, elegant products.
Visit Analyzer
Visit Analyzer is a private-domain B2B sales assistant that processes chat logs and call recordings to generate structured project portraits and H5 analytics dashboards.
ArchUnit Architecture Guard
Protect your Java and Kotlin project structures from architectural drift and erosion with the ArchUnit Architecture Guard from Openclaw Skills.
Job Search — Serbia & Remote (job-search-belgrade)
An intelligent AI coding agent skill that scrapes regional and remote job boards, cross-references findings against your PDF resume, and outputs sorted listings with fit scores directly in your terminal or via SMTP email digests.
Featured*
