Hey, we've created an interactive game generated fully with AI sounds. You can play it @ https://qforge.studio/the-doorstep

There's also a builder if you want to build games like that yourself @ https://qforge.studio. For now it's completely free and no login required so you can just jump in and build. Would love some feedback!

This past year, I've become an avid Claude Code and Codex user. Over that time, I've repeatedly encountered a few frustrations:

* When I provide long, detailed protocols in prompts or CLAUDE.md / AGENTS.md files (e.g. make a plan, implement, test, debug, git commit, etc...) the agent will often skip or handwave steps.

* Often I'll find myself repeating the same patterns of prompts repeatedly. Examples: "diagnose the error" followed by "fix it", looping back and forth between "implement this spec" and "audit the implementation against the spec"

* The agents are fairly limited in terms of scope and max time spent on a per-prompt basis. This makes it challenging to set up long autonomous runs, e.g. overnight.

Today I'm happy to share *FlowCoder*, the project I've been working on to address these issues. FlowCoder allows you to create and execute custom automated workflows for Claude Code and Codex, via a visual flowchart builder. I am hoping this project can both help vibe coders scale their results and enable autonomous agent research by building on top of existing coding agents.

FlowCoder lets you set up slash commands to execute flowcharts of prompts and bash commands. These flowcharts have a fair number of features:

* The core building blocks are Prompt blocks, which send prompts to Claude Code or Codex, and Bash blocks, which run bash commands.

* FlowCoder keeps track of variables while executing flowcharts. Prompt blocks allow you to enforce the agent to respond with structured output to assign variables values, and Bash blocks allow you to save the bash output and/or exit code to variables.

* Branch blocks let you configure a boolean expression with these variables, splitting the flowchart into True and False paths.

* Flowcharts can accept CLI-style string arguments, and all blocks support syntax for argument substituion and variable substitution. So for example, you can create a prompt block that says "Create a spec for this task: $1" and it will substitute the first argument you pass in. README explains more.

* Command blocks allow you to call other slash commands from within your flowchart. FlowCoder maintains a stack of flowcharts to handle command recursion.

* Flowcharts also support Refresh blocks for resetting context and Variable blocks for initializing/setting variables.

* FlowCoder automatically creates a git commit after each Prompt or Bash block.

I've included a number of examples in the repo to help users get acquainted with the system, showcasing prompting paradigms like implement-audit loops and test-fixing loops, and programmatic paradigms like for-loop behavior. README explains more.

Note that these example flowcharts are not "optimized". They are a starting point. Flowcharts provide a huge amount of expressive power. You can encode the specifics of how you like to approach your software engineering practice, whether you prefer to vibe code in small chunks or large autonomous sequences. I have my own set of flowcharts I've been developing for my own practice, and I've seen significant gains as I've been going through the process of optimizing these flowcharts' structures and prompts.

I hope others can benefit from this work or may want to contribute! The project is still very young (v0). The codebase is in alpha and should be assumed to be UNSTABLE. It has been tested on Linux and WSL. Feel free to post any issues you encounter on the GitHub. Currently, I am using this version of FlowCoder to develop the next version of FlowCoder, an Electron-based version with a better-planned architecture and additional features (multi-agent/parallel workflows, CLI, UI improvements).

More info: https://youtu.be/1COOR6UmpsY

Hi hn,

This post is missing from November and December 2025 from the usual whoishiring account [0], is this gone?

I'm aware that the latest posts barely got any comments looking for freelancers and the market sucks, still, I think that those posts are helpful.

- [0] https://news.ycombinator.com/submitted?id=whoishiring

AI coding assistants are productive but sloppy. They produce code that looks right but:

- Imports packages that don't exist - Uses placeholder functions that do nothing - Leaks patterns from JavaScript, Java, Ruby into Python - Leaves behind dead code and duplicates - Uses mutable default arguments

I built sloppylint to catch these "AI slop" patterns before they hit production.

     pip install sloppylint
     sloppylint .

It detects 100+ patterns across categories: - Hallucinated imports (20% of AI imports reference non-existent packages) - Placeholder code (`pass`, `...`, `TODO`) - Wrong-language patterns (.push(), .equals(), .forEach()) - Mutable defaults, bare excepts, dead code

This isn't a replacement for traditional linters - it catches the specific mistakes AI makes that humans wouldn't.

https://github.com/rsionnach/sloppylint

I built it for myself. After a lot of iterations, this is getting some traction.

It helps me organize my personal life.

Still trying to figure out what else it needs to do, and what can be improved.

Please provide feedback!

BinaryStorage is a PHP binary key/value store designed for fast and efficient data access. It supports any PHP serializable data, provides startsWith/contains searches, and includes data compaction to reduce disk usage.

GitHub: https://github.com/olivier-ls/binary-storage-php

Hi HN,

I solo-developed a Python tool that automatically finds and extracts key gameplay moments from long Call of Duty: Black Ops 6 gameplay videos. Saving hours of manual scrubbing through VODs to find highlights, kills, and other events.

The tool does these things:

Ingests long-form gameplay videos.

Applies computer vision to detect (kills, deaths, medals)

Generates raw highlight clips or compilations without needing manual trimming. Output: individual clips or a concatenated compilation

Performs bulk BO6 VODs analysis on a full Twitch channel.

GitHub (work in progress or demo repo): [https://github.com/karimm-ai/NiceShot_AI]

Would definitely love any feedback regarding the tool or monetization methods.

Thanks.

We were using Sprinto for compliance automation for a year or so. But in our recent renewal we ran into a technical issue related to payment therefore I notified their customer success manager about the issue, which they acknowledged. However, someone from their finance team who super privilege access disabled our account as due date had passed. We begged them to enable our account, which the team took almost 24h to respond to and upto 48h to resolve.

As we operate in regulated space, this was nightmare, therefore we decided to let them know that we were going to terminate the service due to breach of trust and started looking for another oem.

However, their team now tells us termination for convenience is not allowed according to their terms of service and that we have to pay the full year contract regardless of whether we use it or not. We believe we terminated due to breach of trust. They have also threaten us to take legal action if we do not pay in next 7 days.

We believe they are not operating in good faith, bullying us into paying for a situation they have created.

We also believe that their regular employee have extended access into data of their customers which honestly makes us uncomfortable to operate with them.

The number we have to pay is around $4000 which I believe is not too much.

Question here is that should we standup against their bullying, malpractice and fight. And is it worth fighting for and pay their contract.

Please advice. We are smaller startup compared to Sprinto which has raised millions of dollars.

I'm curious about how everyone is using AI lately.

- Is every team member using their own editor/CLI?

- Is there any sort of alignment across the team? Like using a shared memory bank for the whole project?

- What about rule files? Does the project have a main rules file, or does everyone just use their own private/local rules file?

I really want to understand the nuances here, what works for you personally versus what works for the team.

Is there an industry standard emerging yet? or do you think people are still trying to figure it out?

I’m asking because it feels like in some workplaces/companies, it's treated like a forbidden topic (which is crazy).