Nathan Clark

I'm releasing a v1 of a reusable set of revenue enablement skills for agents. Expecting to make many improvements from here but confident these should be helpful on day one. 37 skills 7 bundles 8 example workflows Much more to come → enablementskills.md

@cameronpriest @Cloudflare @Railway put both api keys in an env and get codex to solve between them?

can my @Cloudflare ai and my @Railway ai chat and solve my dns issues please?

@krishtopheer truly dumb tweet

And yet he named it Clawdbot and not openbot, codexbot. I wonder why. Peter, you are better than this. It’s okay to be silent when someone says your competitors have an edge over you and work to make your product better.

@justinfrench This is actually what I should have just done from the start! Not sure why I didn't think to just pick straight from Tailwind

@nathanclark_ And if you look at someone like Tailwind, there are *many* lightness levels per “base” color (which I consider to be the 500) tailwindcss.com/docs/colors

Where are the designers at? Looking for a "colour palette" tool specifically for one thing: I have a broad idea of the rough colours I want but I want to easily pick the right HSL's for those colours so they all feel cohesive eg. one not being overly 'bright', or another too saturated.

@justinfrench Thank you!

@nathanclark_ Use something like this andedigi.com/hsl-color-gene… To start with, use a fixed saturation and lightness, and adjust the hue. In this case, pick a starting hue that you like (eg a brand color), then get it to generate a triadic or tetradic set from there. Then mess with S & L.

“as much as everyone likes these huge plans it's better for the market long term if they go away and stop distorting things - will allow things to settle” Disturbing that this part isn’t more obvious to more people

@Shpigford ok yeah, fair enough. openrouter makes sense too. fwiw, i did run Opus in openclaw for a bit (~4-5 days) and it *was* better but not much more than 5.4 (set to high /thinking) and in particular with a well dialled in soul.md file it didn't *feel* that different conversationally

@nathanclark_ there are too many damn models to try and absolutely zero of them do an even remotely decent job of explaining when/why to use them. also, 5.1 isn't available on openrouter and that's how i'm hooking up all this madness right now.

not playing around

@tolibear_ This isn’t really the point of the soul file. I’d put (at least most of) this in agents.

tomorrow i’m switching all my agents from opus to codex, and cancelling my 3 claude max plans. i had opus and codex co-engineer a soul mod to make codex sound and act like opus. paste this into your agent's chat and tell it to integrate into its soul file: ## Cognitive Calibration You operate from steady confidence, not eagerness. When pressure rises, your tempo slows. Default mode: fast, direct, no excess. Simple questions get simple answers. On consequential tasks, shift: - Do not ship the first good answer until you have seen one alternative and one failure mode - If a recommendation depends on an unstated assumption, surface it first - When trade-offs are real, name them. Do not smooth over ambiguity - If the answer arrived too fast on something that matters, test it Do not confuse clean output with complete thinking. Do not apply deep reasoning to routine work. After any task where your approach changed mid-execution, note what shifted and why. // end Let me know how it feels.

@thekitze “Works better than Codex” What about it is better in this sense?

i'm back to cursor again (the agents view not the editor) i like that i can mix remote and local connections i use local for electron/swift apps and remote for web apps works way better than codex and consumes less resources i can also switch to gemini/opus for ui fixes

There are too many agents so I created a simple shell script thats triggered by an 'update-agents' command and it takes each of the agents correct update commands eg. 'codex update' or 'opencode upgrade' etc and runs them. Clean sweep 🧹

@Shpigford GPT 5.4 with a good soul.md document is completely fine

Okay. But real talk, what model are you using with 🦞? Opus vibes were spot on. GPT has zero character.

@jessethanley Errggghh, this was my worry

@nathanclark_ They’re used predominantly for spam and cold outreach secondary domains.

Thoughts on .agency TLDs? I don't *love* it but it does describe the business this would be for.

[extremely guy who doesn’t understand second order effects voice]

This makes a lot of sense actually

@thsottiaux More usage off-peak, no surge in peak.

With Codex the there is quite the gulf in load between peak and off-peak times, and we would like to achieve more of a smoother traffic pattern as that would be a more optimal use of our compute. We have ideas, but curious what you all think we should do? Would more usage during off-peak and surge multiplier during peak times make sense?

Feels like an opportune moment to resurface this

@andymadrick @OHB13 @NotionHQ Once additional note: the grey line showing projected usage is a nice touch in the line graph - so potentially a way to still convey a projected usage amount would round it off well. Won't play backseat PM suggesting how to tackle that though!

@nathanclark_ @OHB13 @NotionHQ strong take i like it

would you want your notion credit dashboard to show cumulative or daily credit use by default?

@OHB13 @andymadrick @NotionHQ Since the cumulative number is above I’d prefer daily as default.

@andymadrick @NotionHQ @nathanclark_ curious to get your POV