Check code against user intent with review agents and tests
A developer has agents re-read all user messages to re-derive intent, then checks code and tests against it, tracking work in a reviewed todo file.
Done with a custom-built or unnamed agent
- What they did
- The developer has the agent re-read every user message posted in a project to derive a fresh statement of what the user wanted, then uses that to check the code and tests. For long work, the agent writes a todo list kept as a project document, recording feedback on what happened as each gate closes. Tests, logging, and interacting with the running app support this.
- How it went
- The developer says they have used this approach for about half a year to cut bugs and keep code matching intent. No bug counts or other measurements are given.
- Worth knowing
- Re-reading all user messages takes only a few minutes even on a project with 10,000 messages, and the task file gets reviewed twice, before and after implementation.
Try it yourself with Claude Code
Read all of my messages in this session and re-derive what I actually wanted, written as a short list of requirements. Then check the code and tests in [folder or branch] against that list and report anything missing, extra, or untested. Record the gaps in a file called todo-review.md and stop there, without editing code until I approve the list.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Oct 16, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Spec features with AI and hand off to coding agent
Author discusses a spec with an AI agent or ChatGPT, then hands it to an agent to code, and tests the outcome instead of reading every line.
Click through macOS app builds to catch visual bugs
A developer used Claude Code computer use to compile, launch, and click through macOS app screens to find layout bugs.
Build a homepage from Figma designs via MCP
The author connected Figma to Cursor over MCP to build a new homepage for a font archive, with great results after Lovable disappointed.
Route Linear tickets to PRs across multiple repos
A developer sends tasks via Telegram/OpenClaw into Linear, where Cyrus routes each to the right repo and model and opens PRs for review.
Update a legacy Telegram bot and refactor an app
The author used Replit Agent to update a Telegram bot to a newer API, which only worked after some iterations, and to refactor a card-collection app including a database schema change.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.