#openclaw #aiagents
#openclaw #aiagents
Still early but already useful for agent workflows.
Install:
npm i -g @claudioabierto/agentkit
Docs
elclaudioabierto.github.io/docs/
Repos
github.com/elclaudioabierto/registry
Would love feedback from people building AI coding agents.
AgentKit changes the model:
Agents install capabilities, not packages.
A capability = a deterministic recipe for a task environment.
Example:
web-automation
media-transcription
PII-redaction
tensorflow-gpu
Each one defines exact steps to provision the environment.
Right now agents treat setup like this:
βTry installing something and hope it works.β
But project environments are not deterministic:
β’ different OS
β’ missing system deps
β’ conflicting packages
β’ undocumented steps
Agents end up burning huge token budgets just guessing.
Most AI coding agents still guess their way through project setup.
install β error
retry β error
retry β maybe works
This wastes tokens, time, and compute.
So I built AgentKit.
A capability provisioning layer for AI coding agents.
(To be continued)
Thank you for sharing!
Excited to share our new paper in PNAS! π
"Identifying air quality monitoring deserts in the United States" www.pnas.org/doi/10.1073/...
Find your county
cascadelab.shinyapps.io/aqmd/
@cascadelab.bsky.social
The University of Virginia has created a website of archived government websitesβbookmark itβyouβre going to need it later.
guides.lib.virginia.edu/c.php?g=1451...
π Introducing tidypollute: an R package designed to simplify working with environmental data. It currently supports data extraction from EPA AirData flat files and the Atmotube Pro API.
@atmotube.com
π Explore the package: lnkd.in/eUgVAn-8
New art challenge:
Draw for an hour a day INSTEAD OF doomscrolling on socials for an hour a day.
Who's in?