Implement a 24/7 security scanning system where a local AI model continuously scans your codebase for issues, and a powerful cloud model periodically reviews and validates the findings for a cost-effective solution.
Source episode · 20:42
How I AI: Alex Finn's Local AI Fleet and Automated Software Factorywith Claire Vo
From 20:42 to 22:24, Alex Finn explains how a local model continuously scans code for security issues and hands credible findings to Claude Code for deeper analysis. Clip range: 20:42 to 22:24.
A prioritized security findings report with validated risks and suggested code fixes.
Anthropic AI assistant
Step by step
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
5 steps
On a machine with high unified memory, such as a Mac Studio, load a large local model like GLM 5.2. Set this model to run as an agent, scanning your codebase for potential security issues at a regular interval, for example, every 30 minutes.
Instruct the local model to compile all potential security issues into a single, structured text file. A Markdown report is a good format because it's readable and easy to parse. This report will be the input for the cloud model.
You are a security analysis agent tasked with scanning a codebase for potential vulnerabilities. Analyze the code provided below. For each potential issue you identify, create a finding that includes: 1. A description of the potential vulnerability. 2. The relevant code snippet. 3. A severity rating (Low, Medium, or High). Format the entire output as a single Markdown file. [paste code to be scanned]
In a powerful cloud-based AI environment like Claude Code, set up a recurring loop to run once a day. The blog post mentions using a command like/loop 24 hours.
Set up a powerful cloud model, such as Claude Code, to run once a day. Its task is to ingest and analyze the Markdown report from the local model, reviewing the entire list of potential issues.
The cloud model acts as the expert 'closer' by reviewing the code snippets for each finding, determining which issues are legitimate security risks, and suggesting how to fix them.
You are an expert security code reviewer. Your task is to prioritize validated security risks and suggest fixes. For the list of security issues below, prioritize them from most to least critical. For each issue, provide a clear explanation of the risk and a corrected code snippet that implements the fix. [paste list of validated security risks]
Turn an idea into a PRD, user stories, and a plan.
After the steps
How to recover when the loop fails and where human judgment helps.
Recover
Keep building
Log your workouts effortlessly without ever opening a spreadsheet. This workflow uses the Claude mobile app to parse a simple voice note and automatically update your workout tracking spreadsheet in the background.
Create a custom email management interface within Claude to read, triage, and draft replies without opening the Gmail tab. This workflow helps you reclaim focus and compound your learnings by training your AI on every interaction.
Build a fully automated system in Claude to generate personalized, interactive HTML proposals and client onboarding materials, saving hours on administrative work and creating a high-touch experience.
Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.