TheAuditor v2.0 – A “Flight Computer” for AI Coding Agents
I’m a former Systems Architect (Cisco/VMware) turned builder in Thailand. TheAuditor v2.0 is a complete architectural rewrite (800+ commits) of the prototype I posted three months ago. The "A-ha" moment for me didn't come from a success; it came from a massive failure. I was trying to use AI to refactor a complex schema change (a foundation change from "Products" to "ProductsVariants"), and due to the scope of it, it failed spectacularly. I realized two things: * Context Collapse: The AI couldn't keep enough files in its context window to understand the full scope of the refactor, so it…
In plain words
TheAuditor v2.0 is a tool designed to help AI coding agents manage complex refactoring tasks by addressing context collapse and stale knowledge issues. It acts as a flight computer for AI-assisted code changes, breaking down large architectural modifications into manageable scopes so AI models can maintain accurate context and avoid hallucinations. The tool is built for developers working with AI on substantial codebase refactors, particularly schema changes or foundation modifications that exceed typical context windows.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I’m a former Systems Architect (Cisco/VMware) turned builder in Thailand. TheAuditor v2.0 is a complete architectural rewrite (800+ commits) of the prototype I posted three months ago. The "A-ha" moment for me didn't come from a success; it came from a massive failure. I was trying to use AI to refactor a complex schema change (a foundation change from "Products" to "ProductsVariants"), and due to the scope of it, it failed spectacularly. I realized two things: * Context Collapse: The AI couldn't keep enough files in its context window to understand the full scope of the refactor, so it started hallucinating, "fixing" superficial issues. If I kept pressing it, it would literally panic and make up problems "so it could fix them," which only resulted in the situation going into a death spiral. That’s the villain origin story of this tool. :D * Stale Knowledge: It kept trying to implement Node 16 patterns in a Node 22 project, or defaulting to obsolete libraries (like glob v7 instead of v11) because its training data was stale. I realized that AI agents are phenomenal at outputting working code, but they have zero understanding of it. They optimize for "making it run at any cost"—often by introducing security holes or technical debt just to bypass an error. This is a funny paradox because when "cornered/forced" to use cutting-edge versions, syntax, and best practices, it has zero issue executing or coding it. However, it’s so hilariously unaware of its surroundings that it will do anything else unless explicitly babysat. I built v2 to be the "Sanity Check" that solves a lot of these issues, and it aims to continue solving more of the same and similar issues I face. Instead of letting the AI guess, TheAuditor indexes the entire codebase into a local SQLite Graph Database. This gives the AI a queryable map of reality, allowing it to verify dependencies and imports without needing to load "all" files into context. A/B Demo: https://www.youtube.com/watch?v=512uqMaZlTg As seen in the demo video, instead of trying to read 10+ full files and/or grepping to make up for the hallucinations, it can now run "aud explain" and get 500 lines of deterministic "facts only" information. It gets just what it needs to see versus reading 10+ files, trying to keep them in context, finding what it was looking for, and trying to remember why it was looking to begin with. I also learned that regex/string/heuristics don't scale at all and are painfully slow (hours vs minutes). I tried the regex-based rules/parsers approach, but they kept failing silently on complex files and suffered constant limitations (the worst offender was having to read all files per set of rules). I scrapped that approach and built a "Triple-Entry Fidelity" system. Now, the tool acts like a ledger: the parser emits a manifest, the DB emits a receipt. If they don't match, the system crashes intentionally. It’s no longer just a scanner; it’s a guardrail. In my daily workflow, I don't let the AI write a line of code until the AI (my choice just happens to be CC/Codex) has run a pre-investigation for whatever problem statement I'm facing at the moment. This ensures it's anchored in facts and not inference assumptions or, worse, hallucinations. With that said, my tool isn't perfect. To support it all, I had to build a pseudo-compiler for Python/JS/TS, and that means preparing extractors for every framework, every syntax—everything, really. Sometimes I don't get it right, and sometimes I just won't have had enough time to build it out to support everything. So, my recommendation is to integrate the tool WITH your AI agent of choice rather than seeing it as a tool for you, the human. I like to use the tool as a "confirm or deny," where the AI runs the tool, verifies in source code, and presents a pre-implementation audit. Based on that audit, I will create an "aud planning." Some of the major milestones in v2.0 * Hybrid…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, December 2025
the whole month →- GP
Life & fun · Dec 2025 · dosaygo-studio.github.io
- JG
Hi everyone! My name's Luke and I made the original Jmail here alongside Riley Walz. We had a ton of friends collaborate on building out more of the app suite last night in lieue of DOJ's "Epstein files" release. Please AMA!
Life & fun · Dec 2025 · jmail.world
PlanEat AI▲733AI turns your health goals into a 7-day menu & grocery list
AI · Dec 2025 · planeatai.com
- 2G
Community, All the HN belong to you. This is an archive of hacker news that fits in your browser. When I made HN Made of Primes I realized I could probably do this offline sqlite/wasm thing with the whole GBs of archive. The whole dataset. So I tried it, and this is it. Have Hacker News on your device. Go to this repo (https://github.com/DOSAYGO-STUDIO/HackerBook): you can download it. Big Query -> ETL -> npx serve docs - that's it. 20 years of HN arguments and beauty, can be yours forever. So they'll never die. Ever. It's the unkillable static archive of HN and it's…
Dev tools · Dec 2025 · hackerbook.dosaygo.com
