Connect with the Product Hunt community

We launched today, and we'd rather you hear the limits from us before you install.

• A prompt you type into Claude Code or Codex is not masked, because neither client lets a hook rewrite it. shim reports it, or blocks it under enforce. Copilot CLI is the exception: there the model gets the masked prompt.

• In Claude Code, the output of a failed command can't be replaced. cat .env && cat missing-file exits non-zero and the model reads the .env as is. shim reports it.

• Files attached with @ and shell commands are reported, not masked.

• Codex and Copilot CLI tool traffic isn't inspected yet. VS Code reports but masks nothing.

• Detection is patterns plus checksums. It misses names, postal addresses and passwords with no key name. Treat it as a guard, not a DLP boundary.

• No Windows yet. macOS and Linux work.

Why a hook and not an MCP server? An MCP server only sees calls to its own tools. The agent's built-in Read and Bash never pass through it. A hook sits on the client's own tool calls, so it sees those too.

Which of these would stop you from using it? Your answers decide what we fix first.

Now you can follow what's new on Oracle Cloud Infrastructure and DigitalOcean, one service at a time on Cloud Releases.

Oracle Cloud Infrastructure: https://www.cloudreleases.io/pro...

DigitalOcean: https://www.cloudreleases.io/pro...

Study with Aura | A Personal Tutor, Whenever You Need One

Hi,

FinderGit 0.54.0 is about the Agents dashboard. It already reads what your coding agents load before they start; now it also finds what they leave behind on disk, helps with a skill listing that has grown past its room, and reads Claude Code's settings, and your organization's policy, the way Claude Code itself does.

What the agents leave behind

Open the terminal where your agent is running. Do it now, before you read the rest.

Is something sitting there waiting for you? How long has it been there?

You re halfway through debugging on a remote host.

A command is still running, but you need another shell to inspect a log. Do you reconnect, pass through the jump host again, and find your working directory all over again?

This small interruption is why I added Duplicate Tab to xtty.

• It opens a new, independent SSH connection to the same host.

• It returns to the same remote directory.

• The original session stays exactly where it is.

Hey Product Hunt

We ve launched Noodle Seed here twice this year, and I wanted to thank this community and everyone who has tried the product and shared feedback with us. You ve helped shape what we re building next.

Across those launches, we learned two lessons. Our first product was easy to get started with, but left too little room for different business needs. Our second gave engineers more flexibility, but getting started still required technical work that many businesses couldn t take on.

Your feedback has helped us focus our roadmap on bringing those qualities together: making Noodle Seed approachable for a business owner / founder, with enough flexibility to adapt as their needs grow.

Hi everyone,

Vitals 1.8 is out. It s the biggest update since launch.

The weekday kept coming out wrong in my agent. Reading the time from the system was never the issue. It would take that time, compute the day on its own, and get it wrong. So now a hook feeds it the current time along with a two week calendar on every message I send. If a reply contains a date that doesn't line up, the agent has to rewrite it.

From then on, I stopped letting it generate anything that a program would read. Frontmatter, JSON, config values, file paths, URLs, dates. A script handles all of those, and it also decides pass or fail. The model's only job is writing the plain text.

When I genuinely can't avoid handing a decision to a model, I don't trust a single one. Gemini will review whatever Claude came up with. For the larger tasks, Codex and Grok get pulled in too.

Are you still letting your agent pick dates or decide pass or fail anywhere?

Over the last year I went from one side project to 7+. Shipping got cheap; keeping an eye on everything didn't. So I built the screen I wanted: every project in one place, sorted by what needs me first.

Today it covers, for every project from one script tag:

A percentage tells you what's gone. Lancetta tells you how long you can keep going like this.

The quick tour of the menu bar, in three reads:

Hi everyone,

Last week I said Comcent was going omnichannel. It's live today.

Your team now answers calls, SMS, web chat and email from one Inbox in the browser. An AI bot can reply first on any channel and hand the conversation to a person when it needs one. Each customer has one contact history, so whoever picks up sees the earlier calls, texts, chats and emails.

The voice side is unchanged: call flows, queues, transfers, recordings, and AI summaries on every call. Pricing is still from $20/month with no per-user fees.

A. "Gorgeous little spot in Silver Lake. Great pastries, lovely place to sit"

B. "Order the 12oz, not the 16oz. Pistachio milk and no sweetener. Go before 9am or you won't get a seat."

Which one actually makes you go? And what is it about a recommendation in general that makes you trust it?

Hi, I m Andrei, the maker of SplitHub. We re launching our coding-agent workflow on October 15.

A coding agent can build three versions of a page quickly. Choosing which version actually helps still needs real traffic and a well-defined conversion goal.

SplitHub gives Claude Code, Cursor and Codex skills, CLI and MCP tools to create experiments, manage them, and read the results. React and Next.js SDKs assign variants locally. The SDK is MIT licensed; self-hosting provides raw counts, while SplitHub Cloud adds statistical analysis and verdicts.

Explore the dashboard without signing in: https://app.splithub.io/demo

Hey everyone, Said here, the maker of CrewTower.

On October 1, Claude Code shipped mods. Hooks could already tell an outside app what Claude was doing and answer a permission prompt. They could never talk back. A mod can.

A mod is a small JavaScript or TypeScript file that Claude Code loads into its own process. It gets called on every event (a tool call, a prompt, a turn ending, part of the screen being drawn) and can watch it, rewrite it, or take it over. It can draw panes and buttons, add slash commands, read context numbers directly, and reach outside Claude Code.

That last part is what I needed. CrewTower lives in the MacBook notch and shows every agent session you run. With a small mod, each Claude card in the notch can now talk to its session:

I m Abbas, the founder of Palance. We build portfolio analytics for people who want to know why a portfolio did what it did, not just what it s worth today. Attribution, sensitivities, simulation, and an AI assistant for when you d rather ask than click.

I m curious how people actually use this stuff. When your portfolio jumps or drops, what do you reach for first? A breakdown of what drove it, a stress test, or a quick look and a deep breath? And what s the question you wish your portfolio tool could answer but can t? The weirder the better. Some of our best ideas started as this is probably too niche and turned out to be what everyone wanted.

And this time, we're talking about a problem that anyone working with multiple AI agents will eventually run into.

Your AGENTS . md says one thing. Your CLAUDE . MD says another. You update a preference in one place, forget about another, and suddenly your agents are working with different versions of the truth.

Hey everyone, Aryan here

A few of you asked for more agents today. Right now Agent Pets works with Claude Code, OpenCode and omp, and Codex, Gemini CLI and Hermes have all come up in the comments.

Which one do you actually use day to day? I'll build the most asked one first.

Hi Product Hunt! AgentDock launched here on 10 October. Since then it has had eleven updates, and the latest one is the biggest: AgentDock now works with many more agents than Claude Code and Codex.

New in 1.3: more agents

• Beyond Claude Code, Codex and Grok, AgentDock now reads Gemini CLI, GitHub Copilot CLI, OpenCode, Kilo Code, Antigravity CLI and Claude Cowork, and shows Cursor's agent under its own name. Their dots show when they're working, waiting for you and done, and their finished turns go into History, Activity and Tokens. Nothing to install.

• They're in preview until each one is checked in a live session: turn them on in Settings Agents. Answering from the Inbox and typing replies stay with Claude Code, Codex and Grok for now.

• Tokens from the other agents get their own series in the hover card and the Tokens widget.

• The full list, and what AgentDock reads for each agent: agentdockapp.com/agents

You asked, it shipped

What started with a simple idea is growing into a community of people who want to move money across borders more easily.

I keep 12 Claude Code subagents in a project's .claude/agents/, one file per role. Each file only decides four things: name, a description saying when to delegate, the allowed tools, and the model.

Two choices I'm unsure about:

This week I went through the sign-up flow of ten well-known products, imagining an AI assistant doing it for me. It could read what each product does and what it costs. On almost every one it would stop at the email confirmation or a CAPTCHA. I had never thought of a sign-up form as something software would fill in.

I see three ways to handle it, and I keep going back and forth:

1. Keep sign-up human-only. Safer, and nobody has to think about bots with accounts.

2. Let the agent do everything except one approval step, where the person confirms the email and accepts the terms.

3. Give agents their own path, like an API key or OAuth login, separate from the human form.

I keep seeing people draw parallels between Obsidian and Make0 Brain. And honestly, I get why.

Both store context. Both can connect to AI agents. Both can help AI work with information you've accumulated over time.

But there's a fundamental difference in how they approach memory.

DungeonQ's goal is to give security teams and authorized defender agents time and space to respond while a designated suspicious session works in a persistent synthetic world.

v0.15.0 adds a reproducible console showing the next step: what happens after an alert?

1. A real MCP client enters the workspace, checks conflicting claims and saves a note that survives restart. Wrong Tickets remain useful only inside their issuing world. A claimed "source restored" message does not change the route. Follow-up work is bounded and offers a stop; participant text cannot authorize operator changes.

2. A separate local receiver acknowledges an alert and rotates a temporary connector credential. Readbacks check that old authority and old-ticket use are denied, authorized new reads work, and those results survive restart.

I recently designed and vibe coded my portfolio. It has an animated intro, project pages, and a guestbook built with React and Firebase.
The site works, but it loads on my laptop is a low bar. If you ve shipped a vibe-coded website, what did you catch in your final review that you missed while building?
I m deciding how to review mine and would rather learn from real mistakes than follow a huge generic checklist.