Brent Traut

In a time where we spend most of our day talking to robots, it's pretty fun to take a little time to share with humans.

I also make my AGENTS.md file and all my skills public. People have already helped me tune those just by asking questions like, "Why do you do that?" or "Should we just push this into the repo?".

It's also embarrassing as hell expose how I talk to robots, especially given that a lot of my conversations expose how little I know about a particular topic. But that's the human condition, and curiosity trumps pride any day.

I'm trying a new experiment: A couple of times a day, I have Codex export all the prompts that I've made across all of my threads, scrub them for secrets, and post them on a web page accessible to all of my coworkers.

The idea is to encourage openness. I want folks to learn from me, and especially for me to learn from others. Hopefully it starts conversations about better prompts, better tooling, and a better future.

@cansar @Dimillian It’d be a sad group of folks getting together, tirelessly trying to improve their own support group.

@Dimillian We should start a support group. It would probably just turn into the above though.

Being passionate about what you’re building is both a blessing and a curse. I can never stop thinking about the next steps. What’s missing. What’s to improve. Reading feedbacks. Always cogitating, never stopping.

@kentcdodds It’s a really great feature. It’s been on Codex for months.

I'm going to use this all the time

@kentcdodds Well, one frontier model and Grok are coming out on the same day.

Two frontier models coming out the same day!?

@_mauriciorubio Will they disappear if I pack ‘em in?

@btraut So you’re playing tetris right? 😆

Things have escalated.

@photomatt @pangram The prompt totally is!

@btraut This is still identified by @pangram as AI-generated.

Since I've been using Codex a lot more to write on my behalf (Slack messages, PR descriptions, docs, etc), I wanted it to get better at using my voice. To best do that, I had it analyze my writing style via existing Slack history, blog posts, and a few hand-written docs, then create a skill that it can use in the future. You can try this as well with this prompt: --- Analyze all Slack messages I've written that are retrievable through my authenticated Slack connection, then create a reusable skill for writing in my voice. Resolve my Slack user ID, then search messages authored by me across public channels, private channels, DMs, and group DMs. Follow every pagination cursor and partition searches by date so Slack's result limits don't truncate the corpus. Read representative threads for context. Exclude copied code, quotes, pasted agent output, automated reports, and bot messages from the voice analysis. Create a detailed voice guide covering my stable traits, context-dependent modes, structure, vocabulary, humor, formatting, and how I ask, disagree, explain, praise, apologize, set boundaries, and handle incidents. Include quantitative evidence, representative excerpts, synthetic examples, confidence levels, and gaps in the corpus. Then create a skill under ~/.agents/skills that: - Triggers whenever an agent intentionally writes on my behalf - Requires reading the full voice guide before every draft - Never invents my opinions, commitments, approvals, deadlines, emotions, conclusions, ownership, or admissions of fault - Treats drafting as separate from permission to send - Does not trigger for normal assistant conversation or unrelated professional artifacts Update ~/.codex/AGENTS.md with one concise rule requiring the skill whenever writing as me. Validate the installed skill and verify that its canonical guide exactly matches the generated analysis. Do not send any messages.

@theo @steipete Mine just showed up and I reheated pizza last night. Theo continues to cost me money.

@steipete 🫡

How did I ever function without AI? cc chefcook @theo

@wwtchr "Why is Codex swearing so much lately?" … "Oh yeah."

@btraut fine-tuned on your own slack history. the model knows you now. whether that's good or bad depends on your slack history.

@PaulSolt @Dimillian @emanueledpt @steipete It’s a bit of a hack, but you can also use emojis (or have your chief of staff use them) to denote thread status. Does it need your attention? Is it waiting on review? Is it parked or blocked? Make it more obvious with emojis!

My NEW Codex workflow is better than I expected. 8 new features ready for release in my app. Took a few attempts to figure out the workflow and some bugs. Feels like the future. Thanks @Dimillian @emanueledpt and @steipete for the inspiration.

@acrognali Costs me 10M tokens a day

its crazy that in the year 2026 there's no public API for adding an image to a github PR

Every time you think you need a dashboard to look at data, stop yourself. Do this instead: 1. Ask your agent to make sure that you have all the data to analyze something actually stored in the database. 2. Ask your agent to write a skill to gather that data. 3. Ask your agent to do the analysis and create a temp and throw-away HTML dashboard to answer the question(s) that you have In my experience, every dashboard that I've created gets less and less use over time and decays. It's much better to make sure your agent can get the data you need and answer the questions you have, on-demand.

Diablo 2 introduced the concept of "Ethereal" which means a weapon otherwise has great stats, but you can't repair it and it dies as soon as it's durability drops to zero. I'm pretty sure Dyson vacuums all have that.

@fractal_friend @SIGKITTEN I’ve some big things cooking!

@btraut @SIGKITTEN Please do a performance blitz this week. It’s quite brutal right now

It's not glorious work, and it certainly doesn't deserve headlines, but I'm still really proud to work on making the app feel a lot more polished and stable. So many people, myself included, live in it all day every day, and we want to do our best for those people.

@Dimillian chief of staff, create a new thread with a /goal to make a /side with subagents that complete the main quest

/goal ask /side to complete the main quest