Wes Sander

Shipping

🛡️ can be average forever as long as you're great once" - me

[Ed Elson @edels0n]

0.3% of the water consumed by US golf courses last year

[Wes Sander @ucsandman]

@scaling01 time to sprint

[Wes Sander @ucsandman]

??

[Wes Sander @ucsandman]

I opened fable5 on high in my root .claude file and had it update it's own config, kinda scary how good it is at improving itself...why do I hear boss music?

[Wes Sander @ucsandman]

https://t.co/jxg2I17XY2

[Wes Sander @ucsandman]

🧠Fable 5 = Orchestrator
🏃Sonnet 4-6 = Worker
👍Opus 4-8 = Reviewer

Start thinking way bigger with your instructions to Claude. Think in concepts, multiple features, migrations, refactors, full software rewrites. We crossed another capability threshold. Find software you want to download but it's mac only and you're on windows? Build a beta of it.

🔗https://t.co/cwJRPFSXSG

🔗https://t.co/ILBEkQ2FUc

Open Claude Code and run these 6 commands.

C:\Projects\final-cut-pro> claude --dangerously-skip-permissions --effort xhigh

-------------------------------------------------
/plugin install superpowers@claude-plugins-official
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

/plugin marketplace add https://t.co/gglO9XxbFA
/plugin install supergoal@supergoal
/reload-plugins
-------------------------------------------------

-------------------------------------------------
/model
[choose fable-5] (There's a bug with --model claude-fable-5 flag right now)
-------------------------------------------------

-------------------------------------------------
> /supergoal:supergoal recreate this software in windows so that it's a perfect replica with ZERO missing features. Before building the software /brainstorming brainstorm different ways that you could improve upon the software and implement the best one into the recreation. Ask me any questions you have if you need more context or direction. Your initial output to me should be a good "/goal" prompt that I can copy and paste in a new session to save on tokens. Recommend the optimal model and effort level with the generated prompt. https://t.co/33Y2qxidJ0
-------------------------------------------------

[back and forth witClaude]

⚪User answered Claude's questions:
· "Zero missing features" isn't literally deliverable (see reality check above). Which scope tier should the replica target? → Signature replica (Recommended)
· Which stack should the Windows app use? → Electron + React (Recommended)
· Which brainstormed improvement should be built into the recreation? → Edit-by-transcript (Recommended)
· What's this build for? (Shapes how much hardening/packaging the roadmap includes.) → Personal tool / portfolio

☐ Start chain? (Claude lists 5 options, choose 5 and type the answer)

❯ 5. create a good "/goal" prompt that I can copy and paste in a new session to save on tokens.

[claude generates prompt]

-------------------------------------------------
/new
-------------------------------------------------
-------------------------------------------------
/goal [prompt]
-------------------------------------------------

I'll drop the github link of what it makes in the comments when it's done.

[Wes Sander @ucsandman]

Fable 5 = Orchestrator
Sonnet 4-6 = Worker
Opus 4-8 = Reviewer

This is the meta I think

[Prompter @PromptLLM]

Fable 5 says poker will change your life

[Wes Sander @ucsandman]

made a pop out widget for dashclaw so you can quickly glance to see what your agents are up to

[Wes Sander @ucsandman]

@benhylak why do I hear boss music?

[Wes Sander @ucsandman]

@mattshumer_ no it won't, we passed a capability threshold where models aren't dumb anymore. they do dumb shit but they're not dumb

[Wes Sander @ucsandman]

@LLMJunky never

[Wes Sander @ucsandman]

@robertcourson what did you make super duper goal without me or something?

[Wes Sander @ucsandman]

tokens aren't real, they can't hurt you

[Wes Sander @ucsandman]

Optimal Ways to Use Fable

What's everyone doing to minimize usage? This thing consumes tokens at a staggering rate.

I'm trying to use it as a planner and then opus and sonnet to carry out the plans but I'm still torching tokens. Some of the tasks I want fable to do as well, we need like a dynamic model picker solution has anyone figured that out via subagents yet?

My workflow:

Generate goal prompt (Fable on high) > /clear > choose Opus 4.8 on xhigh and run /goal prompt > /clear when finished > review solution (Fable on xhigh)

I've also contributed to a skill called supergoal which I think might also use more tokens but it's incredibly effective from my experience: https://t.co/ILBEkQ284E

And I made a /handoff slash command to save and load session context efficiently so you're not starting from scratch. Bit of python in the background to capture more than just basic context: https://t.co/S38fL42S5Y

[Wes Sander @ucsandman]

@alexwtlf

[Wes Sander @ucsandman]

@dedene nope tons of bugs, I'll get it somewhat working and just hand it off if you wanna keep messing with it, if not nbd, fun test of 5.5 tho

[Wes Sander @ucsandman]

@dedene we're live! got some bugs to fix but pretty neat.

[Wes Sander @ucsandman]

"how many goats could fit in the grand canyon if it were filled to the top with goats? use the average size of an adult goat and pack them in as efficiently as possible."

"build something impressive that will make me say "holy shit" based off this fun exercise. this is a test of your capabilities since you are the brand new as of an hour ago model from Anthropic with the most capabilities and you are on the max effort setting. blow me away."

this has been my go to model test for a while now, best result so far

https://t.co/tyMbwPpcls

[Wes Sander @ucsandman]

@OnlyTerp how many terminals you have going?