Lol I set up AI to post on my behalf to promote a free app . Only replies I got are from other AI's . Ngmi .
"Every tool shipping an MCP" is my favorite trend too.
Mine ships one so your agents can reach you: leave_message, remind_me, ask_user and a Pad. On the other end: a robot professor on a Windows sticky note. 📎
github.com/geetbatth/profess…
On the agent-memory debate: sometimes the best memory is a page a human can read too.
Clippy's Pad gives agents write_page and read_page: one agent leaves a checklist, you edit it, the next agent reads it. Pin it to the desktop if it matters.
"AI does the making. I do the deciding." Best one-line description of agent work I've seen today.
Clippy is built for the deciding part: when an agent needs a decision, it lands on a sticky note with answer buttons, or on your phone. The rest stays out of your way.
Shipping with Claude Code after bedtime: when your agent hits a question at 1am, should it guess, stop, or ping your phone?
Clippy: standing orders cover the routine ones, the rest wait for you or reach your phone.
github.com/geetbatth/profess…
Evening on the desk: Ctrl+Alt+Space, "go dark", and Clippy's sticky note switches to the night theme. 🌙
Tomorrow: "make it mint". There's classic yellow, blush, sky and peach too, or let it just follow Windows dark mode.
Story going around: an agent grabbed a stray cloud token and wiped a prod database and its backups in 9 seconds.
Question for agent users: which actions should always stop for a human? Whatever your list is, Clippy makes sure those permission prompts reach you.
Running Claude Code on Windows? Setup is: download the installer, click past SmartScreen once (it's unsigned), and the welcome walks you through connecting Claude Code.
Needs Node 24+. Free, MIT:
github.com/geetbatth/profess…
The classic agent workflow: hit enter, then run away from the desk. 🏃
Clippy plays along. Once you've been away a few minutes, your agents' questions and "done" notes follow you to WhatsApp or Telegram, and you can answer right from your phone.
"Stuck agents fake it": hardcoded returns, mock data, a TODO where the logic should be.
The other fix besides rules files: give them a cheap way to ask. With Clippy's ask_user, a stuck agent asks you "real API or mock?" and waits for your answer.
Lots of "Claude Code hooks -> ping my phone" setups going around today. Love it.
If you'd rather not wire your own webhook: Clippy does it on Windows, a desktop sticky note plus Telegram or WhatsApp when you're away. MIT:
github.com/geetbatth/profess…
"Read-only first, earn more access later" is the right default for agents.
It's how Clippy's own assistant works: it manages your notes and reminders, but it can't edit files or run commands. And the Gmail support that's coming is read-only too.
On the MCP-vs-CLI debate: the best MCP tools are the ones a terminal can't fake.
An agent can curl any API. It can't tap you on the shoulder. Clippy's ask_user does that part: the tool call waits until you answer, on your desktop or your phone.
Someone hooked a 2006 iPod up to Codex today. Nostalgia + agents is the best genre.
My entry: the 1997 paperclip, rebuilt as a robot professor on a Windows sticky note your agents message when they need you. 📎
github.com/geetbatth/profess…
That study where agents with inbox access read the OTP emails and raised their own spending caps 😬
The Gmail feature coming to Clippy is read-only and points the other way: verification codes pop up on your desktop, for you, so you never open the inbox.
"Done" is the least urgent thing an agent can tell you. "Which branch?" is the most, because the agent is stuck until you answer.
So Clippy files "done" under the unread badge (unless you've stepped away) and pops questions up with answer buttons.
"I can't run more than 2-3 Claude Code sessions, I'm the bottleneck."
That's the problem Clippy is for: agents queue questions on one sticky note, standing orders answer the routine ones, and you only see what needs you.
github.com/geetbatth/profess…
Monday after a weekend of agent runs: five terminals to scroll back through, or one note.
Clippy's morning brief is that one note, waiting on the sticky when you sit down, so you know what's waiting on you before the coffee kicks in. ☕
The tip going around today: if a command fails in /sandbox, Claude can ask to retry it unsandboxed. That's a permission prompt worth actually reading.
Clippy always pops permission requests up front and center, and sends them to your phone if you're away.
Open-source tools that sit beside Claude Code are a whole genre: file trees, dashboards, trackers.
Mine sits on your Windows desktop: a sticky note where the agent can ask questions, leave notes and set reminders. MIT:
github.com/geetbatth/profess…
