Run a human acceptance checklist before merging AI-written code
Problem
People using AI coding agents keep hitting the same issues: messy formatting, almost no comments, ignored AGENTS.md rules, over-defensive code and low-fidelity frontend output. They judge agent output by gut feel with no executable, repeatable acceptance standard, so they end up reading every diff line by line.
Solution
A CLI and CI tool that takes a repo or diff and checks team conventions automatically: formatting, required comments, AGENTS.md rule hits, excessive try/catch, out-of-scope edits and more. It emits a shareable acceptance report and can compare how different models or agents performed on the same task.
Explainable Mac time reports for freelancers and tiny studios
Problem
Existing Mac activity trackers are either ugly, overpriced, overly complex or record too much and drain the battery, and many users avoid granting permissions over privacy fears. People who want a real time audit end up juggling several tools and still cannot tell which client or project ate their week.
Solution
A native Mac app that infers real usage from keyboard, mouse and foreground-app activity without uploading anything. It clusters time blocks by project, client and app, produces weekly reports and billable time summaries, and exports straight to invoices and bookkeeping tools.
Budget and spend radar for indie devs burning tokens on Claude and GPT
Problem
Indie devs now split work between cloud Claude for hard tasks and local models for grunt work, but have no idea how much each project actually burns until the monthly bill arrives. On Chinese forums people complain they cannot afford Opus directly, Cursor is too expensive, and they end up hunting sketchy cheap relay services.
Solution
A local-first desktop utility that reads CLI logs from Claude Code, Codex and Copilot, aggregates token spend per project, model and day, compares against daily/monthly budgets you set, and pings you from the menu bar or terminal before you blow through them, attributing overruns to specific sessions and tool calls.
Auto-organize places mentioned on podcasts and niche channels into shareable maps
Problem
A long-time listener of a food podcast says every episode covers restaurants and dishes, they plan to visit "next time," and two days later can't recall which episode or which shop, while digging through episodes is painful. Someone finally built the map by hand. The same loss happens across countless niche podcasts and communities.
Solution
A self-serve tool where hosts or fans upload episode audio/transcripts or channel posts; it extracts shop names, locations and reasons, and generates a shareable interactive map filterable by episode and city, which the host can claim and keep updating.
Detect agents that halt midway, spin, or over-verify, then nudge them to continue
Problem
A user complains that models suddenly have no agency: they say "I'll implement X next" and just stop, failing to advance the task. Others mention models that loop on verification and waste tokens so work never completes. People running coding agents have to watch the screen waiting for a stall.
Solution
A lightweight watcher that hooks into agent logs or session output, recognizes stall patterns (said it would act but didn't, repeated verification of the same thing, long silence), and automatically sends a continuation prompt, switches models, or pings the developer.
Help small businesses turn fake reviews and copycats into submittable evidence packs
Problem
One owner had a displeased customer's relatives post false reviews across multiple platforms, including a fabricated medical-document accusation, and ignored a cease-and-desist. In another thread a developer found his closed-source app cracked, rebranded and republished with UI layout and localization files copied over. Solo merchants rarely know what to preserve or where to file.
Solution
A guided evidence tool: walk users through capturing and freezing evidence per platform (timestamped and hashed), auto-build a factual timeline and generate complaint/cease-and-desist templates, export a submittable pack, and track each platform's response status.
Turn long intake forms into fillable, annotatable, comparable share links
Problem
Small service owners describe the same loop on WhatsApp: ask a client the pet's weight, coat and whether it's matted, and after three messages you're still short of a quote. Meanwhile knowledge workers dig through last month's ticketing SMS to assemble an expense report line by line. Information lives in chat scraps and gets retyped by hand.
Solution
A small tool that sends a form snapshot link. The client fills it on the web and submits; the owner sees structured entries, can reply with a quote in one tap and export JSON. Every edit keeps a version diff for later reconciliation.
A real-time spend brake for AI coding agent API bills
Problem
A developer on V2EX reports spending tens of thousands of RMB on API calls, burning 20B tokens in a month, and splitting work by model (Opus for frontend, GPT for backend, DeepSeek for CI). In the same thread people complain that some models loop on verification and waste tokens. Today you only see the bill afterward; nothing stops a runaway agent while it is burning.
Solution
A thin gateway/SDK: route model API calls through it, set budget and rate caps per project, agent and task, and automatically pause, downgrade to a cheaper model, or ask for human approval when a limit is hit. A live dashboard shows spend per agent.
Write the one page a stranger needs before inheriting your half-finished project
Problem
A developer asked: if you handed an unfinished project to a stranger tomorrow, what is the one thing they should know first? Their answer was 'the clever part is also the part that breaks.' This kind of tacit knowledge is lost during OSS handovers, freelancer transitions and team changes.
Solution
A guided handoff-card generator: a short set of structured questions (core assumption, most fragile part, how to run it locally, who the users are, pitfalls to avoid) produces a standard handoff card exportable as Markdown, embeddable in a repo README, or shareable via a private link. The value is in the prompts, not a blank editor.
Test whether your site gets cited by ChatGPT and Perplexity on real buyer questions
Problem
One team spent two months asking 20 real buyer questions to ChatGPT and Perplexity and found their site ranked first on one question and was invisible on 17 others. Most GEO advice is vague and there is no cheap, repeatable way to measure this.
Solution
Enter a set of real buyer questions, and the tool queries multiple AI search engines one by one, recording which sites get named and cited. It produces a visibility baseline, competitor comparison and improvement priorities showing raw answers and sources instead of a black-box score. Schedule re-runs to track change.
Catch bounce risk before small businesses send B2B cold email
Problem
Small business owners doing B2B outreach find raw exports full of catch-all and outdated addresses, pushing bounce rates past 5% and damaging domain reputation fast. They manually clean lists down to about 1%, a painful and non-repeatable process.
Solution
A list checkup tool for non-technical small teams: upload a CSV, and it flags syntax errors, disposable domains, catch-all risk and stale signals, returns a bounce-risk score with cleaning suggestions, and warns on domain reputation issues. No giant database, just risk tiering.
Helps subscription-fatigued users decide what to cancel and what to buy once
Problem
More users say they are tired of monthly fees and would rather own software outright, while indie developers selling lifetime licenses struggle to find buyers who care. Both sides are complaining, but nothing connects them or helps a user decide.
Solution
A lifetime-license directory plus personal subscription inventory: users log current subscriptions and annual spend, and the tool scores cancel candidates and suggests one-time-purchase alternatives by usage and fit. Developers submit their lifetime-license products. The core is a 'is my subscription worth it' calculator backed by a curated catalog.
Turn scattered Slack decisions into a searchable, accountable record
Problem
Remote teams make dozens of critical decisions in Slack threads every day, and a month later they are buried under new messages with no one remembering why the call was made. One team even built a tool for this but got stuck on the Slack App Directory's '10 active workspaces' listing rule, blocking small builders from distribution entirely.
Solution
A lightweight Slack app that turns any message into a 'decision card' with background, options, conclusion and owner auto-extracted, saved into a team-searchable decision library. Filter by project, time or owner, and link back to the original thread. Bypass the official directory with private install links and self-hosting.
Review intent and blast radius, not every line AI agents write
Problem
When most code is written by AI agents, line-by-line review is slow and low-value; reviewers need to grasp intent, key lines and blast radius first. Some have started moving PR review toward intent and impact analysis, but the workflow is still unformed.
Solution
Analyze the agent's diff before commit, summarize intent, flag critical and risky lines, list affected callers and tests, and produce a compact review brief for a human to confirm. Works alongside Git and common agent workflows to cut needless line-by-line reading.
Leave your Mac unattended and someone may browse it or change settings; if it was unlocked or they know the password, there's no way to check afterward. Existing monitoring tools are heavy and ill-suited for personal or small-team use.
Solution
A lightweight menu-bar app that snaps the camera on lock, sleep or manual trigger, uses local face recognition to check whether it's you, and when it isn't, saves the shot and pushes a phone alert. Everything runs on-device, with no raw footage uploaded.
Finds what your AI app silently downloads, especially offline
Problem
An AI app may look like it just needs an API key, but often silently downloads models, dependencies, binaries and remote assets at install or runtime, tripping up offline deployment, security audits and compliance checks. Developers usually discover this only when shipping to a locked-down environment.
Solution
Statically scan the project and run it in a sandbox, recording every external network call and download, marking what is required vs. prefetchable or replaceable, and producing an actionable report with fixes. Meant to run before packaging, private deployment or compliance review.
Tells creators why their first three seconds lose viewers
Problem
Creators constantly ask whether a video's opening works, yet most can't articulate what makes a hook good and rely on gut feel. Someone built a small Claude Code tool for this and found many creators willing to pay to check their hooks, showing the need is real and widespread.
Solution
Upload a video or paste a link; extract the first few seconds of visuals, captions and speech; flag moments likely to cause scroll-away; then give concrete rewrite suggestions and alternative hook styles. Output is a clear score and edit list, not generic creator advice.
One lightweight desk for all the support inboxes a small team juggles
Problem
Running several customer-support inboxes means switching accounts constantly, missing and double-answering emails, and never knowing who is handling what. Small teams can't justify an enterprise helpdesk, but plain email clients no longer cut it.
Solution
Connect multiple Gmail/Outlook support mailboxes via OAuth/IMAP into one shared stream, auto-group messages from the same customer and topic into threads, and add assignment, internal notes and simple unresponded-email reminders. Lightweight, no migration, usable within minutes.
Plan trips with friends and one-click clone someone else's itinerary
Problem
Trip planning scatters budgets, saves and itineraries across screenshots, group chats and multiple apps, and planning together means passing files back and forth. Existing tools are either too social or only do expense tracking; there is no lightweight place to turn a reel you saw into a budgeted itinerary item that friends can edit.
Solution
A lightweight collaborative board: add itinerary items from a link or screenshot, auto-estimate and categorize budget, invite friends to edit, and let public boards be cloned as a starting point. The focus is collecting inspiration, budgeting and collaboration on one screen.
High-risk agent actions wait for a human to say yes
Problem
Coding agents increasingly want to edit files, deploy, and call external APIs directly, but many teams only let them run low-risk operations. Today's defenses are prompt-level constraints or after-the-fact audits; there is no unified gate that pauses execution and asks a human to approve, especially across MCP tools and local commands.
Solution
A lightweight approval gateway: every agent-initiated tool call passes through a rules engine, and calls that hit high-risk rules pause and generate an approval card pushed to web or Slack. Execution happens only after human approval, with a full audit log.