Try our new dimensional analysis Claude plugin

2026-03-25 · Source: The Trail of Bits Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, short

Summary

A new Claude plugin for dimensional analysis, released on March 25, 2026, assists in developing and auditing code by using Large Language Models (LLMs) to annotate codebases with dimensional types and then mechanically flag mismatches. Unlike typical LLM-based security tools that rely on the model's judgment to find bugs, this plugin employs the LLM as a categorization engine. In testing against real audit findings, it achieved a 93% recall rate with a 12% standard deviation, significantly outperforming baseline prompts that yielded only 50% recall with a 20% standard deviation. The plugin operates through four phases: dimension discovery, dimension annotation, dimension propagation, and dimension validation/triage, generating a `DIMENSIONAL_UNITS.md` file and annotations.

Key takeaway

For developers working on arithmetic-heavy projects like smart contracts or blockchain nodes, running this new dimensional analysis Claude plugin can significantly improve bug detection and codebase understanding. You should commit the generated `DIMENSIONAL_UNITS.md` file and all plugin-created annotations to your repository to enhance both human and LLM comprehension of your project's arithmetic expressions, despite the probabilistic nature of LLMs.

Key insights

This Claude plugin leverages LLMs for code annotation and mechanical flagging of dimensional mismatches, achieving 93% recall.

Principles

• LLMs excel as vocabulary-building/categorization machines.
• Mechanical checks enhance LLM-based security tool reliability.
• Dimensional analysis improves code audit precision.

Method

The plugin follows four phases: dimension discovery to identify base units, dimension annotation to apply anchor annotations, dimension propagation to extend and repair annotations, and dimension validation/triage to discover and classify mismatches.

In practice

• Commit `DIMENSIONAL_UNITS.md` to your development lifecycle.
• Use annotations to improve codebase understanding for humans and LLMs.

Topics

• Claude Plugin
• Dimensional Analysis
• LLM Security
• Code Auditing
• Smart Contracts
• Blockchain Security

Code references

• trailofbits/skills

Best for: Machine Learning Engineer, AI Engineer, Software Engineer, AI Security Engineer

Related on AIssential

• Turning Andrej Karpathy’s LLM Coding Thoughts into Claude.md — LLMs are transforming coding workflows, shifting developers from direct coding to guiding and reviewing AI agents.
• LLM-discovered 0-days — Claude Opus 4.6 finds high-severity 0-day vulnerabilities by reasoning about code, surpassing traditional fuzzing in complex cases.
• LLMs Count with Geometry: Claude’s 6D Helix for Line Breaks — LLMs count characters and manage line breaks using geometric representations in high-dimensional spaces.
• LLMSurgeon: Diagnosing Data Mixture of Large Language Models — LLMSurgeon diagnoses LLM pretraining data mixtures from generated text, enabling post-hoc auditing without direct data access.
• Claude Code experiment: Visualizing Hybrid Logical Clocks — Claude Code, an LLM agent with a terminal interface, streamlines interactive visualization development for complex algorithms like HLC.
• v2.1.56 — Claude Code updates focus on performance, security, and extensibility, enhancing developer workflows and multi-agent capabilities.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Trail of Bits Blog.