Benchmark coding agents on your own PR history—no more gaming public leaderboards.
Problem
Public benchmarks get benchmaxxed fast (Goodhart's law): labs are incentivized to maximize scores, but those scores don't predict whether an agent actually works in your codebase. Developers today rely on scattered trial and error with no trustworthy private evaluation.
Solution
Automatically turn your GitHub repo's historical PRs into real task sets, let coding agents replay them in isolated environments, and score pass rate, time, and divergence in a continuously updating private leaderboard.
One-click cleanup of temp files, logs, and caches left behind by AI coding agents on your Mac.
Problem
AI coding agents leave behind tons of temp files, logs, snapshots, and caches on your local machine; users see disk space shrink but don't know what's safe to delete, and an existing FOSS macOS cleaner proves the pain is real and cross-platform.
Solution
Scan default working directories of popular agents (Cursor, Claude Code, Codex, etc.), identify safely deletable leftovers, group them by type and size, and offer scheduled auto-cleanup plus ignore lists.
A clean, queryable API for secondhand clothing sold prices for resellers and developers.
Problem
eBay's sold data sits behind a login wall, while Poshmark data is messy and often over a year old; resellers either manually page through listings or misprice based on stale comps.
Solution
Provide a query API and lightweight dashboard for sold prices by brand, category, and size, with cleaned, deduplicated data and freshness labels. Developers can embed it in sourcing tools; sellers can see a reasonable listing range at a glance.
Stop reading raw diffs from AI agents—render every PR as an interactive visual map.
Problem
Developers increasingly let coding agents submit PRs automatically, but reading every diff is so tedious that many just merge without review, creating quality risks; someone even built a tool that makes each PR draw itself, showing that 'not reading diffs' is now the norm.
Solution
Generate a visual change map for each AI-submitted PR: highlight risky edits automatically, aggregate touched modules, compare similarity across PRs, and produce a shareable preview page. A quick glance tells you whether the change deserves a closer look.
A bloat-free native task board for small dev teams that plugs into AI agents
Problem
Small teams are dragged down by the bloat and ceremony of Jira, ClickUp, and Trello, and just want a clean, fast tool that doesn't interrupt flow. Existing alternatives are either too thin or are sluggish web wrappers. Meanwhile the team already uses Cursor and Claude, but the task system can't talk to them.
Solution
A native desktop and mobile board that hides advanced settings and keeps only projects, tasks, assignee, and due date. A built-in MCP server lets coding agents read tasks, update status, and generate branch names and commit messages. Data can be local or self-hosted.
A modern mobile client for Navidrome and Subsonic self-hosters
Problem
People running their own music servers often have thousands of carefully tagged tracks and a fine desktop experience, but can't find a mobile client that feels right: existing apps look dated and handle offline poorly. In the Chinese community, people are writing their own Navidrome players, showing this gap persists.
Solution
A modern mobile client that connects to Subsonic/OpenSubsonic: clean UI, smart playlists based on tags and play history, optional crossfade/mixing, and offline caching of albums and playlists so it works without a network. No music hosting, just the player.
Ask a web page a question from the terminal and get only the useful bits
Problem
Developers constantly need one small answer buried in a long page, and browser search or full-page scraping is clumsy. curl-style tools return raw HTML, while AI scrapers dump the whole page into context and waste tokens. Tools like jurl exist because people clearly want curl that actually reads the page.
Solution
A CLI that takes a URL plus a natural-language question, fetches and parses the page, and uses a local or BYOK model to return only the relevant paragraphs, code snippets, and links worth following. Plain text by default, structured JSON for piping.
Deploy approvals enforced in the pipeline, not in Jira
Problem
Too many teams still approve deployments inside Jira or chat, disconnected from the actual pipeline, so someone can just run a deploy and bypass the process. Only after an incident do they realize approvals and releases were never strongly linked, and audits can't tell who approved what.
Solution
A lightweight approval gateway that plugs into CI as a step: before deploy it calls a signed approval API and blocks until approved. Approvals happen via Slack, email, or a one-time web link, and every approve/deny/deploy event lands in a tamper-evident audit log. Integrates with GitHub Actions, GitLab CI, and common PaaS targets.
End-to-end encrypted session handoff for coding agents across devices
Problem
Developers start an agent session on a laptop but want to continue it on a server or hand it to a colleague. Hosted options upload a cloud copy of the repo, so local hooks and config don't come along, and many won't trust a third party with traces worth $50-100 each on the market. Others have ideas on their phone while the session only lives on their desktop.
Solution
A local-first CLI plus PWA: one command discovers and packages the current agent session including local config and hooks, encrypts it client-side, syncs through a self-hosted relay or the user's own object storage, and restores it on another device. One-time share links let colleagues pick up the session without the server ever seeing plaintext.
Early-warning and appeal assistant for IPs silently blacklisted by RBLs
Problem
Indie site owners get caught in ASN-wide blacklists like UCEPROTECT Level 3 even when their own IP did nothing wrong, quietly killing site traffic and email deliverability. These listings rarely notify you, so you only find out when users complain or traffic drops. Appealing means hand-checking dozens of RBL sites one by one.
Solution
A scheduled monitoring service: register your domains/IPs and it checks major RBLs and ASN-level lists daily, alerting you via email or webhook the moment you're listed, with a ready-to-send appeal template per list. A passive-DNS self-check page lets you confirm an ASN-wide false positive in seconds.
Monitor your systemd services at a glance from the desktop tray
Problem
Developers run many systemd services on local or remote servers (databases, queues, custom daemons) but lack a lightweight desktop monitor—they either open a terminal for systemctl status or deploy heavy monitoring stacks.
Solution
A cross-platform desktop tray app that monitors multiple systemd services via SSH or local connection, shows status with color and notifications, and supports one-click restart or log view.
Intercept and approve AI-written emails and messages before they send
Problem
Indie developers use AI to draft emails, client messages, and commit messages, but AI text is easy to spot and sending it under your own name hurts your brand. Rulebooks and humanizer prompts don't work well—a capture mechanism before sending is needed.
Solution
A lightweight middleware that intercepts text copied or exported from AI agents, shows the original, and lets you edit or approve before sending to email, Slack, GitHub, etc. Supports custom rules and shortcuts.
Subscription-free browser transcription and summarization for open-source recorders
Problem
Hardware transcription devices are expensive, and their software subscriptions cost even more. The EarWing open-source project offers cheap hardware (ESP32 + 4 mics), but its software is rough and browser-based transcription/summarization is hard for most users to set up.
Solution
A browser-first web app where users upload audio from EarWing or other recorders, run transcription and summarization locally in the browser, and export or search results—no server, no subscription.
Automate pre-delivery QA for freelance developers handing off client work
Problem
Indie developers doing client work have a mental checklist before delivery—env vars, API keys, build config, license files, leaked secrets—but it's easy to miss something. Tools like Proofkit exist but are rough and need more vertical focus.
Solution
A CLI + web dashboard with pre-built checklist templates for web projects, custom script support, and one-click runs that generate shareable delivery reports for clients.
Version, diff and rollback the prompts and tools your AI agents use
Problem
Indie developers using Claude Code, Codex and similar agents now write code through conversations, with prompts and tool calls scattered across sessions and local files. It's hard to remember which combination worked, and git mixes prompt changes with code changes, making rollback painful.
Solution
A lightweight local-first tool that automatically captures the prompts, tools and model configurations used in each agent session, stores version history in SQLite, and provides a web UI to diff and roll back to a known-good combination.
Guides and tools to safely register, subscribe, and recover AI service accounts
Problem
Users whose Claude account was banned registered a new Google account, but after linking a foreign credit card the Google account got flagged and locked, making the paid Claude unusable. Existing information is scattered, and users don't know how to safely unlink, refund, or appeal.
Solution
Provide a continuously updated guide covering safe Google/Apple account registration, payment method risks, appeal processes, and refund tips; include an interactive checklist to assess account risk level.
SLG players must collect resources at night, time attacks, and guard against raids, making long-term online presence exhausting. Existing game assistants are simple scripts that can't understand complex strategy or alliance coordination.
Solution
Provide a game protocol doc and personal token for players to paste into Claude/ChatGPT, letting AI take over account operations: report plans, write battle commentary, and generate daily reports when offline. The game is designed for AI with built-in state query and action APIs.
Precise rift spawn countdowns and guild coordination for multi-timezone MMO players
Problem
Players of games like Aion 2 constantly calculate rift spawn times across timezones and daylight saving, then coordinate in group chats. Existing Excel sheets and chat bots often miss or lag, causing players to miss rifts or arrive too late.
Solution
A pure web tool showing precise countdowns in server time with advance push notifications. Supports multiple characters, servers, and generates a guild-shareable link for teammates to sync.
Turn any image into a bead pattern with color codes, bead counts, and printable PDF blueprints
Problem
Perler bead crafters who convert images to pixel art still have to manually count grids, match colors, and calculate how many beads to buy. Existing image-to-pixel tools only output a pixel image and don't solve the pre-crafting purchasing and construction preparation.
Solution
Upload a photo, pet, or logo and automatically generate a bead grid, Perler/MARD reference color codes, bead counts per color, and export SVG/PNG/PDF blueprints plus CSV shopping lists. Support adjusting grid width, color count, background, and outline.
Automated technical due diligence reports from Git history
Problem
Technical due diligence for acquisitions takes weeks of manual code review and engineer interviews to assess a codebase's true health. Existing code quality tools only scan source code and miss signals like core developer attrition, abandoned critical modules, and accelerating technical debt.
Solution
Connect a GitHub/GitLab repo and automatically analyze commit frequency, author concentration, module decay, dependency health, and test coverage trends to produce a due diligence summary exportable as PDF. Compress weeks of manual work into minutes while focusing on the risk factors acquirers care about most.