We just open-sourced Paperclip: the orchestration layer for zero-human companies
It's everything you need to run an autonomous business: org charts, goal alignment, task ownership, budgets, agent templates
Just run `npx paperclipai onboard`
github.com/paperclipai/paper…
More 👇
It's real: you can ask Fable 5.2 to create a 90's cereal commercial in Blender and it works and it's good
Paperclip Crunsh: All part of a zero-human breakfast
In this context, canaries are low-value trap assets mixed into your real secrets. You plant crypto private keys holding about $20 alongside sensitive data. Any thief who steals the secrets will almost always drain those funds right away, giving you an instant alert that a breach occurred—just like a canary in a coal mine.
Probably a good idea to put $20 in crypto private keys with your secrets as canaries
If your secrets are compromised almost any hacking agent will steal the funds immediately
196 BILLION TOKENS across 50,000+ @papercliping issues.
Can't wait to unleash what we are building on the world!
Thank you @dotta for transforming the way we build product 🚀
the codex docs on /goal are also helpful for improving your prompting in general:
* define your desired end state
* verified by specific evidence
* preserving constraints
* use these inputs, tools, and boundaries
* after each iteration, here's how to choose the next best action
* when you're blocked here's how to report it
it's simple in concept, but requires discipline to actually answer what you want in each of the 6 areas
---
here's the template from the codex docs:
/goal <desired end state> verified by <specific evidence> while preserving <constraints>. Use <allowed inputs, tools, or boundaries>. Between iterations, <how Codex should choose the next best action>. If blocked or no valid paths remain, <what Codex should report and what would unlock progress>.
looking at pstack by @poteto and it's a goldmine of prompt engineering
even her example prompt is the best template i've seen of the four things you need to give your agent in a single prompt:
* a goal
* a checkable finish condition
* the evidence you want to see
* constraints
I'm always surprised by how short pro's prompts are. It's less about length, but ensuring your hitting those four points.
Here's the prompt:
/poteto-mode fix the duplicate notification after retry.
Reproduce it first. Done means retrying produces exactly one notification and normal delivery still works.
Show the failing and passing evidence. Leave a PR for review.
So simple
define what you want, how to know when you're done, what you want to see, and what you shouldn't do
I guarantee you this isn't obvious to most prompters
here's the sauce github.com/paperclipai/potet…
Really like @papercliping but needed a really nice way to understand which agents need me. Plus it's a really nice totally different view to see your teams working, as if in a virtual office 😁 @dotta
first order: "what broke and how did we fix it?"
second order: "what about our thinking was insufficient and what can we do to ensure that doesn't happen in the future?"
agentic engineering means you focus on the second order
adopt 'blameless engineering' (except the agents are blameless, not you)
but the principles are helpful:
* understand the decision in context
* investigate beyond human error
* learn before judging
* turn learning into concrete improvements
* share knowledge openly
These old human principles are EXACTLY what you need for your agentic work
when you fix a bug, don't just fix the bug - ask your agents to go look at the PR that introduced the bug, and then look at the harness traces that wrote the PR
ask 'why did we make this mistake? what context or rules were missing?'
then turn that into a concrete lint, skill, .mdc - whatever you need to not repeat that behavior
like poteto says, slop is not something that "just happens to you".
you get the slop you allow
being effective with agents means you build tooling to prevent your agents from making the same mistake twice
One hundred thousand Github Stars!!
This makes Paperclip one of the top 150 open-source projects of. all. time.
All computer work is becoming agentic work so you need tools to harness the army of agents at your disposal
We've built 📎 for companies to get a handle on agent sprawl and it's popular is because it's good
Paperclip is multi-player and you can use any model, any harness, any time. It's the layer for every connection, every agent, and every human to collaborate on work
The self hosted version is completely free (and you can sign up for the waitlist for the hosted version)
📎📎📎 ITS TIME TO CLIP 📎 📎📎