Verify external images actually render in your posts, not just lazy-load successfully
Problem
Post-publish checks misjudge lazy-loaded images as failures: before scrolling, image complete is false and naturalWidth is zero, but scrolling makes them appear. Worse, platforms often swap external image URLs for their own proxy, so checking only the original src gives false results.
Solution
A CI/cron-friendly checker: given a list of article URLs, it scrolls each target image into view, waits for a real load, and verifies currentSrc to detect proxy rewriting. It returns per-image status plus the final resolved URL, so you can continuously monitor whether outbound images break or get hijacked.
Record and replay repetitive iPhone Mirroring flows from your Mac, no jailbreak
Problem
Many iPhone-only tasks waste time in finger taps. Developers already record actions inside the Mac iPhone Mirroring window, edit them into a flow, and replay, but building conditionals, retries, and OCR waits by hand is too much friction.
Solution
A visual recorder and player for the iPhone Mirroring window: capture taps, swipes, and typing, export them into an editable flow chart, and support waiting for colors or OCR text, simple conditionals, loops, and automatic skip/retry on failure. Everything runs locally on Mac, no jailbreak or device farm.
Flag silently widened permission boundaries in your PRs before they ship
Problem
Reviewing a PR for authorization changes is nearly impossible by eye. Tweaks hide in custom guards, endpoints, RBAC/ABAC rules, and policy-as-code files, and a miss leads to broken access control or data exposure. Existing tools either lean on LLMs or skip this check entirely.
Solution
A lightweight GitHub Action that statically parses auth-related files and route decorators, then comments on the PR with every place where permission scope was widened, showing a semantic before/after diff. No LLM, fast, self-hostable, turning security review into a checklist instead of a memory game.
Route coding-agent tasks to cheap sub-agents by difficulty and stop burning quota
Problem
Heavy coding-agent users on personal plans complain that quotas drain absurdly fast after switching models, and they can't tell which tasks deserve the expensive flagship. Some already hand-route simple jobs to cheap sub-agents, but there is no lightweight routing layer in their workflow.
Solution
A local proxy/MCP layer that intercepts every request you send to the flagship model and rewrites routing by task type. Simple jobs (boilerplate, tests, scripts) go to cheap models like DeepSeek; hard reasoning stays with the flagship. A dashboard shows quota saved per request, weekly burn curve, and sub-agent hit rate.
Let small trusted groups pool Codex and Claude subscription quotas
Problem
Codex and Claude subscriptions are pricey for individuals but often underused. Some try pooling subscriptions with trusted peers, but lack a trusted, auditable allocation layer and fear quota abuse or account risk.
Solution
A self-hosted quota pool where members join with their own API keys or subscription credentials. The system allocates quotas by preset rules, logs usage, and generates statements. Admins can revoke members or adjust quotas anytime, and requests stay local without a third-party server.
Decide whether a cold email should be sent before it goes out
Problem
Outreach teams blast cold emails with poor results and damaged domain reputation. Some are already using Claude prompts to judge whether an email should be sent at all, signaling a missing quality gate before sending.
Solution
After uploading a recipient list and before sending, evaluate each email on recipient-product fit, personalization, spam triggers, and timing. Provide pass, revise, or skip recommendations with reasons.
Help developers create conference talks that don't look LLM-generated
Problem
LLM-generated conference slides look tidy with infographics and dashboards, but are criticized as soulless. Generating a full deck at once strips personal style and gives speakers less control over pacing.
Solution
Instead of generating a full deck, it suggests opinionated page structures, opening and transition lines, and visual styles based on the talk topic and speaker voice. It outputs an editable HTML skeleton that emphasizes narrative flow over chart stuffing, and the speaker fills in content step by step.
Let developers talk to coding agents in Chinese while keeping English context
Problem
The community recommends communicating with coding agents in English, and some believe using Chinese may increase the risk of Claude account bans. Developers who think in Chinese either translate manually, losing nuance, or use disconnected translation plugins.
Solution
A bidirectional translation layer inside coding agents like Claude Code and Codex: Chinese input becomes idiomatic English prompts, while English replies are translated back to Chinese in real time. Code blocks, file paths, and commands stay untouched, with a glossary for project terms.
Set rules for agent leftovers and keep your project root clean
Problem
Developers delegating simple tasks to DeepSeek V4 Flash and Qwen3.8 Flash find that agents dump scripts and research notes directly into the project root, creating a mess after a few rounds. Claude and GPT tend to behave better, leaving users unsure whether it's the harness or the model.
Solution
Scan the workspace before and after agent runs to detect temporary scripts, notes, and logs, then archive them into a .agent/ directory or prompt for cleanup. A configurable rules file defines what counts as an intermediate artifact and where it should go, with a hygiene report after each session.
A searchable, verifiable directory of MCP tools with call examples
Problem
The MCP ecosystem is exploding, but developers struggle to discover available tools, their capabilities, and how to call them; existing lists are scattered and lack standardized validation and examples.
Solution
An MCP tool directory site: lists tools, search by capability/use-case/model compatibility, shows standardized call examples and schemas, and supports submissions and community voting. Can expand into an online tool-call sandbox for single invocations.
Offload codebase exploration and token-heavy tasks to a local small model to save context and cost
Problem
Using Claude and similar models for code exploration burns tokens on search, file reading, and summarization, which is expensive and pollutes the main context; developers want to delegate mechanical work to cheap or local small models.
Solution
A dev tool/CLI proxy layer: accepts exploration-type prompts (find functions, summarize modules, list dependencies), routes them to a local or cheap small model, and returns condensed results to the main model. Configurable routing rules, result caching, and a clean main context.
A lightweight, customizable always-on-top Spotify lyrics window for macOS
Problem
Users love seeing Spotify lyrics while working, but existing popout windows are intrusive, ugly, or clunky, and there's no option that stays visible without stealing focus.
Solution
A native macOS menu-bar app: fetches lyrics for the currently playing Spotify track and displays them in a draggable, themable, always-on-top window. Supports click-through, font/color customization, and multi-monitor positioning.
A browser-based second opinion on whether an email or link is safe
Problem
Scams and phishing emails are increasingly convincing, and ordinary users (especially older relatives) struggle to tell; enterprise security tooling is overkill and expensive for individuals.
Solution
A browser extension: select any email text or link to get a risk score and plain-language explanation using lightweight heuristics plus public threat intel (URL reputation feeds). One-click report, local history, and a consumer-friendly UX.
A lightweight web editor that lets mountain bikers quickly fix OSM trail ratings
Problem
IMBA ratings and geometry for MTB trails on OSM are often missing or outdated, and contributing via JOSM or iD is too complex for casual riders, leading to errors and abandonment.
Solution
A browser-based lightweight map editor: overlay satellite/lidar/GPX traces, use dedicated keybindings to tag trail segments with IMBA difficulty, and submit OSM changesets in one click. Built-in templates and conflict checks lower the barrier.
Tell a real resolution from one that only sounds finished, so support teams stop paying twice
Problem
A SaaS operator shared that a bot gave a beautifully written reply that was wrong on the one detail that mattered; a human had to reopen the ticket and redo the whole thing. Deflection metrics went up while the team's workload stayed flat, because the reply only sounded finished. Everyone running an AI support layer is asking the same question: how do you tell a real resolution from a fake one?
Solution
A lightweight ticket audit tool: connects to an existing helpdesk or imports exports, cross-checks every AI reply marked as resolved with a second model or human-defined rules, produces a confidence score, and highlights tickets that may have been closed incorrectly; gives support managers a fake-resolution trend chart.
A verifiable independent record of what an AI coding agent actually did
Problem
A developer pointed out that an AI agent's transcript is not ground truth: the agent may claim it edited files or ran tests when it didn't. He built an open-source project, Rashomon, that independently records shell commands, exit codes, tool calls and test results, then compares them against the agent's own account. This kind of agent audit is becoming a real anxiety for teams using tools like Claude Code.
Solution
A personal agent execution recorder: hooks into common agents (e.g. Claude Code), logs the commands actually run, files touched and tests executed, and emits a signable JSON receipt; highlights discrepancies between an agent's self-report and reality; lets users share receipts with teammates or clients in one click.
Keep one muscle memory for people constantly switching between Mac and Windows
Problem
A developer shared that he switches between his Windows PC and MacBook all day, and the most annoying thing is that shortcuts like copy/paste differ across systems, breaking muscle memory and slowing him down. He tried AutoHotkey but found it hard to configure and maintain, which is out of reach for most users. It's a widely tolerated problem with no consumer-friendly tool.
Solution
A graphical shortcut mapper for Windows: ships with macOS shortcut presets (Cmd+C/V/Z etc.) that users can apply in a few clicks; supports per-app overrides, conflict detection and a visual key panel; lets users export configs to share with a team.
Generate side-by-side world events for any historical time window
Problem
Someone shared that when reading history books he struggles to place events in a broader timeline: he knows the dates but not what else was happening elsewhere in the world at the same time. He built Meanwhile Map around that need, which confirms the demand is real, but a standalone web map isn't ideal for quick reading-time reference or for saving and sharing.
Solution
A lightweight contemporaneous-brief tool: enter a year or range and get structured cards of key events happening in different regions and domains; one-click export as an image or short PDF for reading notes; users can save interesting windows into a personal timeline.
A self-hosted inbox filter for freelancers and small teams drowning in cold sales emails
Problem
A developer shared that he was flooded with cold sales emails daily and ended up writing a script that scans Gmail every ten minutes, blocking obvious pitches with a set of regexes and falling back to AI for the rest. Many people don't want to hand their inbox content to a third-party SaaS, but also can't build and maintain these rules themselves.
Solution
A self-hostable email filtering assistant: ships with a continuously updated library of cold-email regexes and phrasing patterns that users import into Gmail/Outlook rules with one click; calls the user's own LLM API key for borderline cases; and provides a dashboard to review blocked mail and tune rules, with all data staying in the user's own account.
Turn any document into validated structured JSON ready for downstream workflows.
Problem
Businesses still drown in invoices, contracts, and forms that must be manually transcribed, which is slow and error-prone. An existing product already does 'any document to JSON,' proving broad demand, but small teams and vertical use cases lack a lightweight, embeddable option.
Solution
An API and web app that ingests documents, extracts fields automatically, and returns JSON with custom validation rules and field mapping so output can be consumed directly by downstream systems; self-hosting for sensitive industries.