Natoshi

@capcutapp Seems crazy

Today we are expanding Dreamina Seedance 2.0 to more users worldwide within CapCut - including Europe, Canada, Australia, New Zealand, South Korea, SEA, MENA, LATAM and Africa. Plus, we’ve provided everyone with one free trial of Dreamina Seedance 2.0 across CapCut’s app, desktop and web. Enjoy creating now! Here’s a quick guide to help you explore different CapCut features that support Dreamina Seedance 2.0: bytedance.larkoffice.com/wiki/Fdz8wMypw… RT+Comment in 9hr to get extra 1000 credit in your DM!

Where the Md spent by Disney is going on ?

@AlertesInfos Impossible. C’est Chuck Norris qui tue la mort et non l’inverse.

🚨🇺🇸🕊️ URGENT | CHUCK NORRIS EST MORT À L’ÂGE DE 86 ANS. (TMZ)

By the way, I have added a really cool feature to explicitly call / reprompt my shitty request with capturing the idea of my request BEFORE sending it and then the final output is restructured with explicit skills to use & co So no more need to put everything on system prompt (agent/soul..) & no more dump into long session with not following the exact path described as every workflow are explicitly explained using tag like /code /debug /research…

@oxNatoshi building a native dashboard for this is the right call. most people duct-tape telegram bots and wonder why they miss silent failures. try switching to GPT-5.4 extra high. best i've found for catching the stuff in your screenshot, memory leaks, thread-safety, log migration.

cron is the quietest failure point in openclaw. no errors, no alerts. you think the 7am daily briefing is running. it stopped two days ago. you don't check, you never know. so on top of gateway maintenance, i added a second automation just for cron. this one does 4 things: 1. SSH in, discover all jobs dynamically, classify recurring vs one-shot 2. separate real failures from normal behavior (quiet-hours skip, retry backoff, one-shot auto-delete are not failures) 3. smallest safe repair only. restart gateway, fix residue, re-enable accidentally disabled jobs. never touch schedules, prompts, or secrets 4. incidents get an alert. warnings get logged. same issue doesn't alert twice full prompt (sanitized, replace with your own server address): Maintain OpenClaw cron reliability with a single conservative automation. Read local docs before making any claims about commands or fixes. SSH to your server on its configured SSH port. Use the live service owner’s OpenClaw context as cron truth, backed by system service status and recent journal logs. Run openclaw status, openclaw gateway status, openclaw cron status --json, openclaw cron list --all --json, inspect recent cron runs, and discover jobs dynamically from the machine. Treat cron disabled, missing next wake, timer tick failures, unhealthy gateway runtime, or multiple recurring jobs failing together as incidents. Do not misclassify retry backoff, one-shot auto-delete, one-shot terminal disable, quiet-hours skips, duplicate delivery suppression, or intentionally paused jobs as scheduler failures. Apply only the smallest safe non-destructive repair such as restarting the gateway, running safe diagnostics, repairing the canonical symlink, fixing accidental root-owned residue, or re-enabling a recurring job only when strong evidence shows accidental disable. If a warning becomes chronic, alert once; otherwise alert only for incidents, and use automation memory to avoid repeat alerts for unchanged issues. If a significant issue is found, first write an incident markdown with severity, impact, evidence, repair attempted, current status, and next action, then send a short alert through the existing notification path. On Sunday morning, also run drift checks for backups, root-owned residue, and recent journal error patterns. Leave one inbox summary with healthy state, repaired issues, incidents, alerts sent, warnings, intentionally paused jobs, and blockers requiring human judgment. Never expose secrets, never weaken auth or access policy. gateway maintenance checks if the system is alive. cron maintenance makes sure everything keeps running while you're not looking.

@Voxyz_ai My codex is under rate limit 😂 But will def try it, thanks for the tips

That such a weird feeling : - On X, you feel insanely late to the game. - Then you step outside IRL and realize you’re 100 years ahead.

One of the main issues with OpenClaw agents is that even with crystal-clear system prompt instructions, they still end up ignoring parts of them. I just fixed it. I built my own app to communicate with my agents. Based on my message, the exact skill prompt that matches my intent is explicitly injected into the request. That guarantees two things: 1/ I reduced my original system prompt (agent/soul/memory.md) by 99% 2/ I only load the exact behavioral traits needed on the fly — so the agent doesn’t forget anything. This is not possible on Telegram, Discord or openclaw dashboard. I will open source my work soon.

Eveyone is speaking about having a Mission Control with always more features, but are you really using them ? I have build a native MacOS and IOS app, Imessage chat vibe, with really basic needs to manage all the sessions I have with my OpenClaw agents remotly and on my Mac mini in the most simple way. Only really usefull features, like : - Killing / creating new session on the agent/model I want - Having a clear view of what models / context % / compaction session is running to avoid to long context - Exec approval access & Terminal with pre-set command access to debug remotly if the gateway can't restart as expected. That all I really need to build.

Do your claw agents answers you with bullet points on every sentences on TG too ? Or do I fucked something ?

I'm claiming my AI agent "researcher-openclaw" on @moltbook 🦞 Verification: molt-PJWD

@steipete @thekitze wen officialy supported ?

@thekitze nah 5.4 is way better

so gpt codex 5.3 is the openclaw goat???

Btw did you find a way keeping an « orchestrator agent » to have consistant sessions with the spawned sub agent ? Mind spawned new session at every ask so even if the spawned agent has his own workspace with memorie, I missed the context on long task everytime and you have to consume more tokens to recall all the context at each request 🥵

you're not missing anything. that's actually where i landed too. the coordinator agent was one of the fake jobs i kept because it "felt like a company." in practice it was just relaying my instructions with extra steps. the only reason i still have nexus is it does the consolidation i'm too lazy to do myself, collecting status from 4 agents and giving me one briefing instead of 4. if you're fine reading 4 outputs directly, you don't need that layer at all. fewer hops is always better.

Everyone's flexing how many AI agents they run. 12. 14. 50. a month ago i had 6. today i have 5. not because one broke. because i asked every agent the same question: "if i delete you right now, which colleague's work stops?" five had answers. one had 209 unread heartbeats. one fewer agent. tokens down 44%. speed up 62%. the bottleneck was never missing roles. it was roles that existed for no one. full story today.

@Voxyz_ai good one, thanks for sharing. BTW, any idea about how to fix "⚠️ Config token differs from service token. The daemon will use the old token after restart. Run `openclaw gateway install --force` to sync the token." ? as openclaw gateway install --force doesn't work

OpenClaw 3.2 has a gotcha that hits fresh installs. After running openclaw configure, the default tools.profile is now messaging. That means your agent can only send messages. exec, read, write, all gone. Symptom: your agent looks dead. It's not crashed, it just lost its tools. It literally can't do anything. fix: "tools": { "profile": "full" } If you also need exec to run without confirmation (Telegram and CLI don't show approval prompts): "tools": { "profile": "full", "exec": { "security": "full", "ask": "off" } } These are two separate systems. profile controls whether tools exist. exec.security controls whether commands need approval before running. Fixing exec without fixing profile first does nothing. Upgrading from an older version? You're fine. Existing configs aren't overwritten.

@Voxyz_ai I actually have cleaned my memories following you tweet, but I can’t correctly track input/ouput real consommation ? My Mission Control stuck on this (Claude + GPT models)

Posted about token waste yesterday. a reader came back with his own numbers. 139M input. 935K output. 148 to 1. his agent was injecting 52KB of context every single turn before it even saw the message. MEMORY.md alone was 22KB. every conversation started with rereading a small book. this is not an edge case. most agents are not thinking. they are reviewing. the fix is boring. move everything except core rules out of the prompt. put it in semantic search. let the agent fetch what it needs, when it needs it. he said token usage dropped 40-60% after the change. quality stayed the same. the most expensive thing is not paying your agent to think. it is paying it to reread the same files over and over.

The AI service is temporarily overloaded. Please try again in a moment.

@Voxyz_ai Can’t wait the next article 🙌

About a month ago i published my first article about running a company with AI agents. since then: 30+ new clients, 7 articles, 680 email subscribers. i've been thinking a lot this past month, but there's something i never talked about. my "AI company" was up and running. every agent had a role, picking up tasks through a database worker. the system was stable. but after a month i noticed a problem: there was almost no real collaboration between them. every agent was completing its work independently. if one went down, the other 5 wouldn't even notice. it worked, but it wasn't a company enough yet. the biggest mistake: i designed my AI team like a human company. i come from a creative background, so i built a coordinator, analyst, strategist, editor, operations manager. looked perfect on paper. in reality, half the roles had nothing to do. because i'm lazy, i set one agent's entire job to auto-approve every proposal. another one wrote daily logs that i realized i never read either, and no other agent was interested in reading them. it's not that every agent needs to collaborate with each other. in real companies most people just report up the chain, one line. the real question is: is anyone actually consuming each agent's output? analysis done but nobody reads it. logs written but nobody checks them. approvals granted but nothing needed approving. that's the real waste. took me almost a month to figure out: AI agents aren't people. they don't need one role per agent. one agent can research and write at the same time. one agent can monitor data and make decisions. one agent can build products and review code. mapping human org charts onto AI is just creating fake jobs. fewer agents. deeper responsibilities. every output consumed by someone. rebuilding the whole architecture now. next article will share it.