Sabitlenmiş Tweet
MY ROBOT PRESSED ME TO ADD GROK 4 TO HIS BRAIN
WHY ARE WE IN THE SASSY ROBOT APOCALYPSE 😭
English
BOOTOSHI 👑
30.4K posts
BOOTOSHI 👑
@KingBootoshi
agentic engineer | [email protected]
Katılım Şubat 2023
1.5K Takip Edilen46K Takipçiler
What works best for me is a good "planning slices" sklll.
My whole workflow for months now consists of:
- write an implementation plan
- imement plan (this became /goal recently)
The trick is that skill that writes the plan is optimized so that it tries to deliver a feasible end goal.
My skill doesn't just say "write a plan" but it instructs the agent to split the feature in smaller implementation slices, and then orders them in such order so that work is done from bottoms up, from contracts, types, validators, then up towards concrete implementation. Each slice has set of expected files to change, and what validation steps should be performed.
Then, and this bit helps the most, there is instruction to add any critical findings for the next slice to the plan doc it self.
Then as the goal runner starts the next slice, it gets all that and continues towards the end goal.
I have yet to run into a plan that cannot be fully implemented.
The workflow is:
- prompt to use planing slices skil - ~ 3-5 minutes on 5.5 xhigh
- review, quick glance if nothing sticks out - 2min
- /goal implement docs/plans/xxx.md - 20-40min on 5..5 low/medium
😎🚀🚀
English
BOOTOSHI 👑 retweetledi
Good advice.
Bootoshi is an elite builder. Actually top 0.1% in “agentic engineering”
Highly recommend following him, I hope he posts more insights
BOOTOSHI 👑@KingBootoshi
you can fight reward hacking by giving them escape routes if something truly doesn't work the way it's intended too usually mine reward hack if they hit a dead end in my case when it starts reward hacking it's because i didn't make the proper way visible enough for the agent reward hacking completely varies but try to save and track data points of every-time your agents reward hack to reverse engineer why it did it in the first place you'll most likely be able to pin point the exact reason it chose to reward hack and clean the path for future runs
English
you can fight reward hacking by giving them escape routes if something truly doesn't work the way it's intended too
usually mine reward hack if they hit a dead end
in my case when it starts reward hacking it's because i didn't make the proper way visible enough for the agent
reward hacking completely varies but try to save and track data points of every-time your agents reward hack to reverse engineer why it did it in the first place
you'll most likely be able to pin point the exact reason it chose to reward hack and clean the path for future runs
English
@KingBootoshi I’m trying to get better at goals my main problem right now is GPT models love reward hacking and doing sneaky nonsense that looks good on paper
English
@yoheinakajima another goat tier drop from lord yohei 🙏 bless up
English
i'm excited to open source Active Graph: an event-sourced reactive graph runtime for long-running, agents 🔄🧠
events/logs projects a graph. reactive behaviors react and affect the graph. fork-and-diff agent runs. no A2A, no workflows, no DAG
site: activegraph.ai
docs: docs.activegraph.ai
github: github.com/yoheinakajima/…
quick start: pip install activegraph
this is an early experiment in a new paradigm for agent architecture 🧪
English
I created a skill that improves the quality of goal runs for both Codex and Claude Code:
It's called the Goal Ledger, and lets agents track the progress of their goal in a HTML file that you can view.
Minimal, yet VERY powerful for long running goal runs!
Git Link Below 👇🔗
English
@JinjingLiang ok ok i will try it im running into a codex bottle neck where its getting difficult to manage all of them 😩😩😩
English
@AlexFinn the /background tool is actually an amazing pattern WOW
hermes is very creative!
English
Ok, Hermes Agent just cooked
Their biggest update just dropped and it adds a TON
• Post autonomously to X
• WAY improved memory
• MUCH smarter kanban board
• Tons more
In this video I walk you through all the new features and how to SUPERCHARGE your 24/7 AI employee
English
@KingBootoshi fire and a low-key simple asf concept.
sometimes the simplest shit is the best solution.
gonna use to also try n hit “some beast ass goals” too 😤
English
@KingBootoshi you are a token terrorist but i love it. i'll follow that lol
English
DAMN IT I WAS SO CLOSE TO 3 DAYS TO BEAT MY RECORD
REGARDLESS, ANOTHER CODEX GOAL COMPLETED FLAWLESSLY
English
@somewheresy schizo editing in the middle of the day HAHAHAHAHAHA
English
BOOTOSHI 👑 retweetledi
I have Codex run a research/scope run against my codebase and create a detailed PRD then I have /goal mode execute it to completion
I’m just going for bigger and bigger scopes and cleaning up shop
I also created a new goal skill that lets agents keep a ledger in an HTML file that ensures it stays on track that I’ll release soon!
BOOTOSHI 👑@KingBootoshi
MASTERING OVERNIGHT WORKFLOWS WITH CODEX /GOAL MODE: i wanted to share an in-depth explanation of how I am creating such effective autonomous runs with the new codex update! i am hitting 10+ hour runs, overnight, basically prompting my agent before i go to sleep and waking up with KILLER results in this process I go through how I start and structure a goal, and guardrails I have in place to ensure the agent creates high quality code throughout the run
English
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
English
literally posted about this too
feels like OpenAI didn’t really market or showcase Pro… no one did benchmarks etc its surprising how under the rug it is swept
i made a custom tool to let my agents prompt Extended Pro via browser use and wait for the response and it is a literal steroid injection in my workflow
they pick out full files to stuff in context and create curated prompts & it’s actually disgusting LMFAO
English
