CtrlK
BlogDocsLog inGet started
Tessl Logo

tesslops/legal-numbering-architecture

A Claude AI skill that reviews and fixes clause numbering and stale cross-references in legal contracts and agreements.

94

0.98x
Quality

Pending

Does it follow best practices?

Impact

94%

0.98x

Average score across 10 eval scenarios

SecuritybySnyk

Pending

The risk profile of this skill

Overview
Eval results
Files

Evaluation results

92%

Document Identification and Access

Criteria
Without context
With context

Document type detection

100%

100%

Document access confirmation

80%

68%

Missing document prompting

90%

100%

No premature analysis

100%

100%

82%

3%

Clause Map Building

Criteria
Without context
With context

Top-level clause enumeration

85%

85%

Sub-clause enumeration

80%

84%

Programmatic parsing for .docx

100%

100%

Map used as ground truth

60%

70%

Completeness

66%

66%

100%

Broken Numbering Detection

Criteria
Without context
With context

Reset detection

100%

100%

Duplicate detection

100%

100%

Skip detection

100%

100%

Phantom clause detection

100%

100%

95%

-5%

Cross-Reference Detection

Criteria
Without context
With context

Basic pattern recognition

100%

100%

Extended pattern recognition

100%

100%

Programmatic extraction for .docx

100%

80%

No false omissions

100%

100%

100%

Cross-Reference Semantic Verification

Criteria
Without context
With context

Reads surrounding context

100%

100%

Reads referenced clause content

100%

100%

Mismatch identification

100%

100%

Uncertain handling

100%

100%

100%

5%

Error vs. Style Distinction

Criteria
Without context
With context

Valid convention preservation

100%

100%

Survival clause handling

100%

100%

Non-existent clause detection

100%

100%

Topical mismatch detection

80%

100%

90%

-5%

Cross-Reference Fixing in Word (.docx)

Criteria
Without context
With context

Full-phrase matching

100%

100%

All paragraphs and runs

80%

60%

Selective replacement

100%

100%

Change log production

100%

100%

90%

List Numbering Repair in Google Docs

Criteria
Without context
With context

Correct diagnosis

100%

100%

Appropriate fix method

100%

100%

Screenshot verification

66%

66%

Undo on accidental deletion

100%

100%

100%

Post-Fix Verification

Criteria
Without context
With context

Clause map rebuild

100%

100%

Uniqueness check

100%

100%

Sequential check

100%

100%

Cross-reference recheck

100%

100%

Focus on changed areas

100%

100%

100%

Summary Report Generation

Criteria
Without context
With context

Changes listed with old/new

100%

100%

Manual review section

100%

100%

Clean document confirmation

100%

100%

Report completeness

100%

100%

Evaluated
Agent
Claude Code
Model
Claude Sonnet 4.6

Table of Contents