Hermann

GitGuardian flagged a leaked Discord webhook from one of my repos. I'd never hardcoded it, so I had no idea where it came from. Traced it to my AI agent's memory syncing notes to a GitHub repo that wasn't private. Audit what your agents write to disk. #buildinpublic #security

My dock went from 20+ apps to 13 in a year. Each one quietly replaced by an AI skill or a terminal session. The bottom of my screen used to be full. Now it's a small island. Didn't expect to feel sentimental about app icons. #AI #mac

@theo all of these agents require shaping and molding, this is just your path to doing that. if your filter for not using it is "it needs to be perfect for me" you're going to struggle to find anything that fits.

Hermes Agent comes with a truly absurd number of skills pre-enabled. Over 100 of them. This is roughly half. I get what they're going for - they want an agent that comes "ready out of the box". I just don't get why every user has to have a polymarket skill, 3 baoyu art skills (? never heard of this), a headless Pokemon skill, and Minecraft modpack server skills, all available the first time they run it. I guess Hermes Agent just isn't for me.

@cowboylexx Oh, cheers man, this is great. Where's the painkillers when I need them?

I have a side project with numbers small enough that I did not want to tweet them. That flinch is the most useful feedback I have had in months. People try it, most never pay, and they are not confused. They just do not need it enough yet. #buildinpublic

@AsifSyedTech I think it's more about just being comfortable in a situation that feels very uncomfortable for me.

@dhlotter That flinch is so real. It’s like a gut check. Sometimes the numbers tell you more than the hype. Maybe it's about finding the right audience, not just the product.

Convex returns non-JSON bodies on 400 schema-validation errors. My naive `json.loads(resp.body)` was crashing without ever logging the response. Now I log the raw bytes before parsing. Half of debugging APIs is reading what they actually return. #convex

@Jbm_dev circumstances change. finally in a place where i can do something like this again.

@Jbm_dev what a nice toy and spot to wrench. i use to have a bmw r80 back in the days.

how do you relax as a founder? I’m trying to watch a movie can’t stop thinking about the product can’t stop thinking about building

@NousResearch still training mine to actually understand my setup. tool search helps, but it really comes down to two things: the tools you give it and the context you feed it. the rest is config.

Hermes Agent now has Tool Search, so your agent only loads what it needs

@OpenAI this scales all the way down, not just to terence tao. i've started attempting things i'd never have touched, reverse-engineering my solar inverter's modbus, scoping a land cruiser EV conversion, purely because the friction to even try dropped to near zero.

AI can give researchers the freedom to pursue “crazier” ideas. For Terence Tao, AI creates more room to experiment, test unexpected paths, and discover what might otherwise stay out of reach.

@gregisenberg mine's turned into a background service. a cron harvests every claude code session into my vault each night, summarizes + wikilinks it. vault gets richer, claude's context gets better, the whole thing compounds. 4.8 sharpened it again.

How to use Obsidian with Claude in 61 seconds

@TTrimoreau honestly the model barely matters anymore. top 3 LLMs all spit out similar frontend. the difference is the skills and prompts you've curated around them, not which one you pick.

Which AI do you use for frontend?

@mikefutia built basically this for my own sites with a few claude skills. great for technical + core web vitals. but it doesn't touch backlinks or domain rating, and that's the bit you're actually paying ahrefs for. not really a replacement.

I just built a complete SEO audit plugin in Claude Code that replaces your $200/mo Ahrefs subscription 🤯 One Claude Plugin audits any store: technical SEO, product schema, content, Core Web Vitals, and AI-search readiness. Parallel agents, a 0-100 score, and a dashboard that renders right in the panel. All inside Claude Code. So I pointed it at Ridge .com, one of the sharpest DTC operators out there. It came back 56/100, and what stood out wasn't a knock on them at all: Ridge has a better AI-commerce setup than 99% of stores. A real llms.txt, an agent-discovery sitemap, a live MCP endpoint, genuinely ahead of the curve. And even on a store that dialed-in, the audit surfaced fixable gaps in ~90 seconds: → Room to add product structured data → A mobile Core Web Vitals score worth tightening → A thin meta description on a high-traffic collection Perfect for e-comm operators and SEO agencies who are sick of paying $200/mo for tools that bury the real issues, running quarterly audits that take a week, and shipping reports nobody can act on. So I put together the full playbook to build your own. The complete guide to building this Plugin in Claude Code: branded to you, tuned to exactly how you audit, repeatable across every client. The kind of audit you run in minutes and hand over as a deliverable that looks like it cost thousands. What's inside: → The architecture (orchestrator + parallel sub-agents) → How to fetch any store past Cloudflare → The 0-100 scoring + falsifiable-findings framework → How to ship the HTML dashboard for client demos → The full build, start to finish Want the playbook for free? > Like this post > Comment "SEO" And I'll send it over (must be following so I can DM)

@marclou no-brainer at the same price imo. faster, better, moved everything over this morning. "back on top" almost undersells it when the cost didn't even move.

I made a way to keep the Ship or Die crew accountable 🏴‍☠️ 1. Connect your GitHub to a mission 2. Unlock trophies for actually shipping 🏆 1st commit 🏆 30-day streaks 🏆 10 commits in a day 🏆 1,000 commits One-shot by Opus 4.8.

@theo Been running it on a gnarly agent-infra codebase all morning. What stands out isn't raw capability, it's that it stops to ask permission less and nails multi-file refactors first try. Benchmarks won't show that, but it's what actually saves my day.

Opus 4.8 is here. It's pretty good. Is Anthropic back on top?

1/ A user messaged me at 2pm: they'd paid for credits, none had landed. Bug on my end. Within 30 minutes I'd found it, fixed it, emailed them it was sorted, and doubled the credits they bought to say sorry. They opened the email and ran 4 generations in 8 minutes.

The official Deye app rounds everything to neat numbers and updates over the cloud every sixty seconds. Useful enough to glance at, not useful enough to actually understand. For months I've been wanting more — register-level numbers, sub-second resolution, the kind of view that lets you actually see the inverter thinking. This weekend it got archaeological. Modbus register maps, V5 Solarman frames, byte-level offsets. Now it talks to me directly over the LAN — battery SoC, panel watts, grid draw, all streaming locally with no cloud round-trip. Watching the numbers shift through the day while I work is strangely grounding. The inverter feels like part of the house now, not part of someone else's server. #solar #homeautomation

SEO Sunday on Nymly this weekend and I was about an hour into setting up IndexNow — the key file, the docs, the protocol handshake, all of it. The kind of work where you can feel yourself becoming an expert on something you'd really rather not be an expert on. Then I noticed Cloudflare Crawler Hints already does the whole thing. One toggle in the dashboard. Free. Signed automatically. Half of what I call "my infrastructure" turns out to be reinventing things Cloudflare ships by default. Spent more time learning the protocol than the toggle would have saved me. #cloudflare #seo

Job-hunting at the Director level is its own kind of doomscroll. The good roles hide behind the wrong title, the wrong company description, three pages of "AI prompt engineer at a crypto startup" before anything real. I was losing an hour a day to the search itself, and not even productively — half-distracted, half-resentful that the search needed me. So the search doesn't need me anymore. A small scanner watches a curated list of companies and pings my Discord at 7am with anything new. Scores fit, dedupes, flags borderline ones. Quietly hunting in the background while I keep building. That feels closer to how this should have been all along. #careerops #automation