Golden Prompts
A prompt is βgoldenβ when it was followed by at least 10 uninterrupted autonomous turns from the agent β no corrections, no course changes, just work. These are the things you said right before Claude Code nailed it on its own. Treat them as reusable templates, or read them to learn what kind of framing gets you leverage.
ultrathink through how to decomplect and then make the complete plan and then execute it if gemini approves
re: pass 3 the qui thing should be fixed the way other secrets work in .env, the Cap repo has some custom build stuff check docker compose to understand. i think you actually can do everything, do it
make sure this is installed (latest version) https://github.com/anthropics/skills/tree/main/skills/skill-creator
Oh, I thought irt was in your context from the last compaction. can you search for "Model B" in the .bak claude code session log?
these are different? Disk Fillingperegrine Firing Disk /media/windows on peregrine predicted full within 30 days (-349GB predicted remaining) for 6h 5m Disk Fillingperegrine Firing Disk / on peregrine predicted full within 30 days (-138GB predicted remaining) for 6h 5m
make sure developers link goes to https://dev.opendatalabs.com
Can you write up a CMS problem brief, without prescribing decisions?as .md file
hey what files in the git repo shouldn't be there?
weird that sometimes i see browser chrome and sometimes i dont
but what are the metrics that can't easily be goodharted?
push the autoresearch md report in ~/code to Simon and have him email it to me
iI rebooted abdand did stuff now itsit's lovkedlocked up again
can you hand this off to another agent? I need you to document everything in an .md file to get them up to speed
just hold on, coordinator is going to send message
jfc write me a new slack prompt that ACTUALLY makes sense
Fixed. The PR description on #98 (https://github.com/vana-com/vana-connect/pull/98) is now clean and readable. so the agent says. get gemini to review it, fix it up if needed, get it merged and released. gemini 3.1 pro not 2.5
can u include comments that i can quickly reference about how to do what i want
oh nvm i misunderstood what those files were. can you look at them though
just keep going and asking yourself: is this *arr quality level?
would like it to show ended (if ended) / different scores e.g. metacritic imdb RT idk / genre (e.g. i wanna be able to view horror so i can wipe em out)
Break making money as an etsy store owner who makes painted oyster shells while working during the daty as a software developer and married with a baby down using first principles thinking. Start by identifying every assumption people commonly make about this topic. Then strip each assumption away and ask: what is fundamentally, provably true here? Rebuild the concept from only what remains. Show me what changes when you remove inherited thinking.
can u write a memo as a deletemeafteryoureadthis....md file to another agent about everything you learned
remind me what we did that we should test with end users?
shouldnlt say what to look for in each source. the list of sources is unknown and will grow over time. skill can use examples. but should not assume too much
why would it loop? i think the trigger should be decided elsewhere. the point is th identify and execute on the next prompt
just for now. pnpm dlx @opendatalabs/connect@canary
lay out implementation plan, sanity check it against important research, then execute it (use sub-agents as appropriate), through canary release
That worked. Is that a potential conflict with other commands? And does anyone on reddit use it?
vana-connect-starter main $!? via π₯ v1.2.19 > ls -la ~/.devcontainer-homes/*/code 2>&1 lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 11 15:07 /home/tnunamak/.devcontainer-homes/clawmeter/code -> /home/tnunamak/code lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 11 15:07 /home/tnunamak/.devcontainer-homes/data-connect/code -> /home/tnunamak/code lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 11 15:07 /home/tnunamak/.devcontainer-homes/dotfiles/code -> /home/tnunamak/code lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 11 10:38 /home/tnunamak/.devcontainer-homes/macro-cart/code -> /home/tnunamak/code lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 11 19:01 /home/tnunamak/.devcontainer-homes/vana-product-interrogator/code -> /home/tnunamak/code /home/tnunamak/.devcontainer-homes/data-connectors/code: total 12 drwxr-xr-x 3 tnunamak tnunamak 4096 Mar 12 10:08 . drwxrwxr-x 24 tnunamak tnunamak 4096 Mar 12 15:18 .. lrwxrwxrwx 1 tnunamak tnunamak 19 Mar 12 10:08 code -> /home/tnunamak/code drwxr-xr-x 3 tnunamak tnunamak 4096 Mar 6 00:00 macro-cart
there are some unstaged changes? did we forget?
yeah i think you need dig into this a bit more and now that yo useem to better understand the scope of what i want, continue working until you've really hit the mark, no question
what ? i thought u were making a connector to get my data?
specifically https://raw.githubusercontent.com/vana-com/data-connectors/69954479eda58737fddb9ade4555d41f36c0c77a/reports/TEMPLATE.md
now please document your run in the .md file for tracking those
Get the latest vana-connect skill and tell me what you'd do differently
can you write a memo for this run? there's a doc we're collecting them into
where did you get your playwright-runner from?
cant nav to product from orders. only from products table. something broke
keyboard shortcuts trigger doesnt work. also navigate/ triggers are full of stuff thats easy to get to/find in the UI itself, and missing things like Empty Search which shows an image. what are ALL of the states in specific views that we need to add? I want to reveal 100% of the app through the design panel. do you understand what i mean by that?
if there is anymore you can do overnight related to this or other GTM activities, make a list, plan them out, do them all, it can take very long that's okay. i suggest defining the long list of validation criteria up front before you begin to make sure you dont stop early
do you see any issues with your implementation? be critical and honest
u dont need to verify inside dev container
we dont need to use trixie if its creating complications. but if it's the *ideal* solution, let's use it. wdyt is the ideal solution?
new window in kitty shows 1 package(s) upgradable sudo nala upgrade Disk usage: 72% --clean will free ~17.6G: docker system prune -af: 17.6G Manual: Conda envs: 27.7G conda env list Devcontainer homes: 75.5G du -sh ~/.devcontainer-homes/* .git bloat: vana-gotchi-python-api (1.0G) git gc --prune=now .git bloat: macro-cart (1.2G) git gc --prune=now Steam games: 105.8G Tool updates: β claude-code-history-viewer 1.5.3 β 1.6.0 gh release download v1.6.0 --repo jhlee0409/claude-code-history-viewer --pattern '*.deb' --dir /tmp/ --clobber && sudo dpkg -i /tmp/*.deb (base) tnunamak@peregrine:~$ node --version v25.7.0 (base) tnunamak@peregrine:~$
fix it you have access to everything dont wait for me
let's extract this app to the root and get it deployed on vercel. are we in a branch for it?
re: the font, what choice would stripe make here?
something has been quietly growing in the past few days idk what
check for unstaged changes that ar eready to commit and deploy, and ensure they deploy to vercel successfully
impractical do do as percentile of population? just asking
i think i worked on this separately, maybe, can you investigate and see? and we shouldn't be capping.
can you investigate why the products table in the macrocart database hasnt been recording upc?
did you check /projects/data-connectors like I explicitly told you to?
why is the whole foods logo in my /project/data-connect just a white W on black? does it need to have some code updates for the logo too? and is there a PR for that?
please dont stop, don't talk to me, keep polling/communicating with the agent who commands you. if you stop, you essentially die because nothing will get you started again
EIP-191 sign has an (i) but nothing happens when i interact with it i think you need 3-10 more rounds
Yeah but remember to documenting important findings, ensure the spec is fully updated, the implementation plan is complete and there is no more "this is a lot of scope, let's only do a phase 0" and we actually go after everything slowly, but surely, and then execute. Any questions?
I upgraded your reasoning effort to max.
overnight: continue working with the other agent, challenge it to do better. be critical.
I'm wondering if this should include tabs or something for standard mermaid diagrams like sequence diagrams (rendered, with copyable text). also the diagram we have should be pan + zoomable and the organization of the boxes feels kind of random?