Check architecture consistency of AI-generated code to prevent agents from introducing bad dependencies or patterns.
Problem
With the rise of AI coding agents (like Cursor, Claude), generated code snippets may violate existing architectural conventions, such as circular dependencies, wrong layer calls, or inconsistent naming. Current linters focus on syntax and style, lacking architecture-level checks.
Solution
A configurable architecture linter that lets developers define project layers, allowed dependency directions, module boundaries, etc., then automatically scans AI-generated changes and warns about violations before commit. Runs as a CI step or editor plugin.
Automatically clean up leftover sessions from AI coding agents like Claude Code to keep your workspace tidy.
Problem
When using scheduled routines in Claude Code, each run leaves behind a session, and the desktop app only allows deleting them one at a time. This quickly accumulates hundreds of useless sessions — e.g., a single routine had 618 open sessions — making manual cleanup extremely time-consuming.
Solution
A lightweight CLI tool or editor plugin that scans agent session directories, identifies sessions created by scheduled tasks, and batch-deletes them according to rules (e.g., keep newest N, protect sessions you replied to). It works via command line or simple config, eliminating the need for manual GUI clicks.
Help developers port large games or desktop apps to the browser with AI-assisted asset shrinking and WASM
Problem
Porting a 1GB+ game like Slay the Spire 2 to the web is bottlenecked by asset size, load speed and browser compatibility. Regular developers lack a systematic way to use AI compression, on-demand loading and WASM compilation.
Solution
A toolchain and guide for porters that analyzes asset size, auto-compresses textures and audio, suggests on-demand loading, includes WASM build templates, and bundles an AI Q&A for specific porting questions to output a runnable web project skeleton.
Centralize AI coding subscription usage in one local dashboard before you burn a week's quota in a day
Problem
Heavy AI coding users complain that normal workloads consume 50% of a weekly quota in a single day, and even $20 Pro plans feel insufficient. When quotas reset or change unexpectedly, users have no unified view to break usage down by task or agent and only notice after the fact.
Solution
A local menu-bar/tray app that aggregates usage from major AI coding subscriptions, shows consumption curves by hour, task and agent process, sets threshold alerts, records quota reset events, and suggests remaining task budgets.
Track your yearly AI workflow experiments and see which skills the models have already absorbed
Problem
AI users' stacks change fast: prompt engineering two years ago, agent workflows last year, MCP and Skills this year. But hard-won usage experience quickly becomes a default model capability, making it hard to see which personal investments still hold long-term value.
Solution
A local-first tool where users log the AI tools, tricks and workflows they rely on each quarter, auto-generate a personal AI skill timeline, and cross-reference model changelogs to flag which practices are now native and which remain worth keeping.
Turn coding-agent wait time into a quick, dismissible microgame or micro-workout
Problem
Developers now run multiple coding agents in parallel and have idle attention while waiting, often drifting to phones or staring blankly. There is no short, interruptible way to kill time inside the terminal workflow. Early explorers built Minesweeper and push-up mods, but no reusable generic solution exists.
Solution
A mod/plugin for major coding agents like Claude Code and Codex that surfaces mini-games, micro-workout prompts or breathing exercises while the agent runs, tracks daily wait time, and returns automatically when the task finishes.
One command to inspect all your VPS for IP cleanliness, streaming unlock and DNS leaks, with AI-ready structured results
Problem
People with several VPSs manually run scattered scripts before renewal or after IP changes to check IP reputation, native IP status, DNS leaks, streaming unlock and latency, then read lines of raw output. Even with an open-source script, there is no polished product for smoother, repeated use.
Solution
A desktop/CLI tool that SSHes into multiple VPSs, runs 10+ network health checks in one shot, produces a comparative table and structured export, plus a built-in prompt that users can paste into any major AI for a plain-language summary and renewal advice.
One menu bar click to route Downloads files where they belong
Problem
The macOS Downloads folder is the default dumping ground for screenshots, installers, PDFs and invoices, and it gets messy fast. The Show HN author built a lightweight menu bar app specifically for Downloads triage, confirming this is a daily, recurring annoyance.
Solution
A menu bar app that moves files out of Downloads based on extension, filename pattern or source app, with 'keep last N days', per-project tags and batch rename, plus an undoable recent-actions log so misclassification is recoverable.
Turn any URL into clean Markdown that AI agents can actually use
Problem
When feeding web pages to AI agents, copying raw HTML brings in navigation, ads and scripts that waste context and hurt comprehension. The Show HN pitches 'turn any URL into clean Markdown for AI agents,' confirming this is a recurring friction in agent workflows.
Solution
A thin, focused converter: paste a URL, extract the main article, strip ads and nav, preserve heading hierarchy, code blocks and tables, and output clean Markdown with optional link footnotes. Ship as a browser extension, CLI and API so it can plug into agent pipelines.
One inbox for every mention of your brand or project
Problem
Indie developers and marketers are spread across Hacker News, Reddit, X, V2EX and Product Hunt, and it's easy to miss when someone mentions their product — losing the chance to reply or do PR. The original Show HN literally pitches 'one inbox for all mentions of your brand or projects,' confirming the pain is real.
Solution
Users add keywords and sites, and the product polls public search results and RSS on a schedule, deduplicates, tags, and funnels everything into a unified inbox with read/assign states, CSV export, and a daily email digest. Start with low-friction sources like HN, Reddit, V2EX and Product Hunt.
See what your Mac's USB-C ports, cables and docks actually do
Problem
Mac users constantly wonder whether a cable really supports 4K 60Hz, whether a dock is silently draining power, or why an external drive is slow — and they end up swapping cables and ports by trial and error. The author behind this Show HN frames it exactly that way: users want real measurements instead of guesses.
Solution
A macOS menu bar app that enumerates devices on each USB-C / Thunderbolt port, reports negotiated PD profiles, actual link bandwidth and Alt Mode support, then runs a short loopback or throughput test to summarize what this port + cable combo can actually do. Results can be exported as a shareable diagnostic card.
Unifies token, quota and cost data from 14 AI coding services into one local desktop view.
Problem
Developers who use multiple AI coding services cannot see token history, remaining quota, available cost and limits in one place; they have to log into each vendor console separately. An open-source desktop app already exists, but cross-platform support and alerting still leave room, and many just want a lightweight local dashboard with reminders.
Solution
A local-first desktop app that reads or imports usage and quota data from multiple AI coding services, shows unified token history, cost estimates and remaining limits, and notifies the user when quota is low or promos expire. Data stays on-device with CSV export.
Snap the cup sticker and collect your daily coffee pickup numbers into a 000-999 album.
Problem
People who buy Luckin or Starbucks daily glance at the pickup number on the cup sticker. It is a free daily number lottery, but once the cup is thrown away the number is gone, and no app helps users keep them. Coffee apps only handle ordering, not collecting.
Solution
Use the phone camera to photograph the cup sticker, OCR the brand and pickup number, automatically cut the number sticker out of the photo, straighten it and add a white border, then light up the matching cell in a 000-999 album. Every cell must be unlocked by a real photo, never typed in; support multiple brand sticker templates and album sharing.
A niche job board purely for hypermedia-driven development roles.
Problem
It is surprisingly hard to search mainstream job boards for hypermedia-driven development jobs; both employers and candidates get buried under generic frontend listings. Someone already built a board just for this, which shows the vertical demand is real.
Solution
A niche job board for the hypermedia / htmx ecosystem: employers post free or cheap, jobs are filterable by stack tags, and candidates can subscribe via email or RSS. A semi-automated pipeline scrapes and dedupes postings from existing boards, with human review to keep it vertical.
Helps people juggling multiple AI plans see monthly spend and decide what to keep or drop.
Problem
Heavy AI users often hold OpenAI team seats, Claude Pro, regional coding plans, Kimi packages and annual Grok plans at once, piling up hundreds of dollars a month. Renewals are messy and promo expiry is easy to forget. Existing tools cover a single vendor or a single coding tool, not a cross-vendor subscription ledger.
Solution
Users enter or paste each subscription (name, price, cycle, promo deadline, seats). The system normalizes everything into a monthly cost, classifies it as keep/watch/cancel, and reminds before promos expire. It can also suggest which plan to open in a given month using existing usage data, and works without any vendor API keys.
Turns a retail trader's written rules into a daily scan-generated buy/sell list and keeps them honest.
Problem
Many retail traders buy and sell on the mood of the day, never look back at the rules they wrote down, and cannot tell after the fact which trades followed the plan and which did not. Broker apps and quote tools only help you place orders; they don't enforce discipline.
Solution
Users visually compose their own conditions (RSI, moving averages, volume ratio, PE, etc.), the system scans a stock pool after market close and lists buy/sell signals, backtests on historical daily bars, and tags every real trade as 'followed the rules' or 'didn't'. No broker connection, no money handling, no stock recommendations.
A predictable-priced time tracking and invoicing tool for freelancers and small teams burned by post-acquisition price hikes.
Problem
Freelancers and small teams relying on time-and-invoicing tools have seen renewal prices explode after acquisitions, with one user's bill jumping from $12/month to about $180/month, while usage-based pricing makes monthly cost hard to predict. What they want is a simple per-seat alternative.
Solution
A lightweight time tracking and invoicing SaaS with a flat per-seat fee, no hidden usage charges, invoice generation for freelancers and small teams, project time logs, and report exports. It offers one-click import from popular incumbents to cut migration friction.
Tells you whether an AI-generated clip is usable before editing, with a use/trim/retry/swap verdict.
Problem
People generating short video assets from text or images often find the result looks good at first glance but fails in the edit: a cup handle changes in the second half, or a camera move blocks the space reserved for subtitles. Without a systematic check they keep reimporting into an editor, burning time and credits.
Solution
A lightweight web tool where you upload or paste an AI-generated clip and first write three lines of intent: what the shot must express, which single motion matters, and which detail must not break. It samples beginning, middle, and end, flags changes in subjects, objects, text, and safe areas, then returns a verdict of use-as-is, trim, retry with one changed condition, or swap, with one line of reasoning.
Keeps coding-agent status, music, and system alerts in the MacBook notch so you stay informed without breaking flow.
Problem
Running several AI coding agents alongside writing and music means constant window switching to check who is working, who is stuck, and what is playing. The MacBook notch sits unused, while existing notch tools cover only music or a single feature and miss the agent workflow entirely.
Solution
A lightweight workspace around the macOS notch and external-display top bar: it aggregates coding-agent run states, attention-needed alerts, music controls, and a focus timer, and lets you jump back to the right terminal or window with one click. It is deliberately read-only and non-intrusive, not a replacement for existing agent tools.
Find a spoken line, a shot, or on-screen text across hundreds of local videos with one sentence.
Problem
People who work with video remember a quote from an interview, a shot they captured, or roughly what was in a clip, but cannot recall which file or minute it was in; they end up opening files one by one and scrubbing. Cloud services are a poor fit for private footage, and manual tagging never happens.
Solution
A local-first desktop tool: point it at folders of video, audio, and images, let it index once, then search by natural language for spoken lines, general meaning, visual shots, and on-screen text such as slides or recordings. Results jump to the exact moment with a preview, and source files never leave the machine.