Created an agent skill called “Visual Explainer” + set of complementary slash commands aimed to reduce my cognitive debt so the agent can explain complex things as rich HTML pages. The skill includes reference templates and a CSS pattern library so output stays consistently
posted similar thoughts on linkedin a few months ago after diving into deterministic authorization world at eqtylab.
I sort of coined the auth model here as "dynamic delegation" where the chain of actions for a task shapes itself & unlocks via human approval. Google's CaMeL paper still offers interesting insight and in ways this might look like code mode in execution.
I think `verifiable intent` is an interesting trust model and doesn't change how this all looks, except all behavior can be cryptographically linked to human intent & verified downstream; intent carried with an identifiable agent.
my team got some key pieces in, like enabling hardware based signatures in a PR in github.com/cloudflare/web…
How my /goals look atm.
The process ultimately spits out the Intent with relevant docs made in the shaping process. I /copy it and start the goal with it. I run the same process over and over (manual loop) for each slice of work - trying to build a thing to automate that.
@arb8020 Ah, ty for the feedback. The review landing is confusing at first. Something to work on.
it’s a cli under the hood, but used as a skill or slash command enables the agent feedback integration.
looks like something formal is coming. This is where there is a big MCP > cli advantage.
There's a lot of "agent native" talk and slopware out there, mostly a facade of forcing an agent to use a cli to poll an api for this type of thing; but the ergonomics are painful. Better to just be _naturally_ integrated.
x.com/mxstbr/status/…
We need a protocol for pushing events into agents. I don't want it to poll, I don't want it to monitor.
MCP doesn't really support this right now, right?
I can hack together something through hooks where you can inject data whenever there's an opportunity, but that's not as
We need a protocol for pushing events into agents. I don't want it to poll, I don't want it to monitor.
MCP doesn't really support this right now, right?
I can hack together something through hooks where you can inject data whenever there's an opportunity, but that's not as fun. The protocol would allow the agent to act, even if it's idle.
I'm still not sold on auto memory. It's good when it works and it's terrible when it doesn't.
I call it a dark pattern because I think it interferes with AI literacy and AIUX - resulting in users not knowing what the hell their AI is seeing or what's actually steering their sessions.
Even for common users - I would say it enables forms of sycophancy for my business colleagues in destructive ways; creating a lot of bias in knowledge work.
We've yet to hone in on the most simple way to do memory (I think I have something here). Everybody creates something sophisticated or let's the model decide - I'm not really interested in either
2K Followers 754 FollowingBuilding https://t.co/0ahzINKrRY to help founders and engineers fix technical support fast. Previously sold 3 startups; here for the love of bootstrapping indie SaaS.
270 Followers 4K FollowingAll trades must have stop loss. Do not fade the market. Do not add onto losers. Add onto winners. Ride the trend -- don’t guess the end. Learn to let go.
1K Followers 2K FollowingProduct of progressive public policy; raised by public libraries and public education that produced a passion for politics. and apparently alliteration
288 Followers 193 FollowingSoftware engineer, building things with and for AI, sharing what I learn
https://t.co/N6XgyFgUBH - lone worker safety
https://t.co/PXpb9o00KL - video planning
52K Followers 374 FollowingPlugins, Apps, MCPs at @OpenAI. Prev Director of Engineering @Shopify, co-founder and CEO @Stellate (acq. by Shopify), co-creator of styled-components.
162 Followers 4K FollowingCo-founder & CTO Gurupass · fitness marketplace 🇧🇷
Construindo produto e dado nos dois lados difíceis: academia e quem treina.
202 Followers 574 FollowingI make emerging tech boring. Battery Pre-Development & Powertrain Innovation @MercedesBenz. Building stuff with AI daily. Efficiency is a personality trait.
288 Followers 193 FollowingSoftware engineer, building things with and for AI, sharing what I learn
https://t.co/N6XgyFgUBH - lone worker safety
https://t.co/PXpb9o00KL - video planning
1K Followers 2K FollowingProduct of progressive public policy; raised by public libraries and public education that produced a passion for politics. and apparently alliteration
2K Followers 754 FollowingBuilding https://t.co/0ahzINKrRY to help founders and engineers fix technical support fast. Previously sold 3 startups; here for the love of bootstrapping indie SaaS.
34K Followers 4K FollowingSay hi, I'm friendly! Founder/CEO @infinite_red, game dev @jammingames (Gunship Origins, published by MicroProse), rec hockey goalie, Lutheran, dad x4, gpa x1