Super User Daily: March 26, 2026
Analyzing 9,358 tweets from March 24, 2026. After filtering: 6,110 quality posts. Top 500 reviewed.
@claudeai [Claude Code]
https://x.com/claudeai/status/2036503582166393240
New in Claude Code: auto mode. Instead of approving every file write and bash command, or skipping permissions entirely, auto mode lets Claude make permission decisions on your behalf. Safeguards check each action before it runs. Available on Team plan in research preview.
@minchoi [Claude Code]
https://x.com/minchoi/status/2036279719386100126
This Claude Code project setup guide has gone mega viral — 2,500+ likes, 520K+ impressions. A comprehensive walkthrough of CLAUDE.md configuration, skills, hooks, and the practices that make Claude Code dramatically more effective.
@iuditg [Claude Code]
https://x.com/iuditg/status/2036337404676809205
Used Claude Code to make an actual stock exit trade today. Combined Claude Code with the /autoresearch skill to analyze positions and execute. 52K impressions. The line between AI-assisted analysis and AI-assisted trading is blurring fast.
@trq212 [Claude Code]
https://x.com/trq212/status/2036442891346755787
Figma MCP update makes it one of the strongest integrations with Claude Code. You can now use Claude Code to design in Figma with full context of your design systems — components, spacing, colors, everything. Backend devs can now ship UI without a designer.
@bcherny [Claude Code]
https://x.com/bcherny/status/2036286592089022699
Boris Cherny revealed the Anthropic Labs team (the team he joined to lead) shipped MCP, Skills, Claude Desktop, and Claude Code. Just a few people, shipping fast. Today they released full computer use in Cowork and Claude Code. The early Sonnet 3.6 prototypes felt clunky — now it controls your entire desktop.
@kr0der [Claude Code]
https://x.com/kr0der/status/2036235321780621738
Discovered Claude Code has an unreleased feature called Auto-dream under /memory. According to Reddit, it periodically runs a subagent to consolidate memory files for better long-term storage. Basically how humans store memories by sleeping — the AI equivalent of dreaming.
@omarsar0 [Claude Code]
https://x.com/omarsar0/status/2036514241167909125
Claude Code now supports auto mode. 8,200+ impressions. The permission model is evolving from binary (approve all vs skip all) toward intelligent, context-aware decisions.
@MillieMarconnni [Claude Code]
https://x.com/MillieMarconnni/status/2036363493478375797
Someone built a Claude Code skill that scans Reddit and X from the last 30 days on any topic, then writes copy-paste-ready prompts based on what the community has actually figured out. Type /last30days and get the top patterns real users are using right now. 100% open source, MIT license.
@binance [OpenClaw]
https://x.com/binance/status/2036390795343962458
Binance AI Pro goes live — a full AI trading agent powered by OpenClaw. One-click activation, multiple AI models (ChatGPT, Claude, Qwen, Kimi), dedicated AI account with isolated API key, spot and perp orders plus on-chain queries. Beta at $9.99/month. The first major exchange to ship an OpenClaw-powered product.
@AlexFinn [Claude Code + OpenClaw]
https://x.com/AlexFinn/status/2036236966757539862
Detailed comparison: Claude can now control your desktop and it works well. But it is still Anthropic's closed garden. OpenClaw runs local models, SSHes into servers, creates its own skills. Claude will never let you leverage other models or run locally. Claude Code is the best way to write code — but not a general purpose, open, self-improving agent.
@aakashgupta [Claude Code + OpenClaw]
https://x.com/aakashgupta/status/2036233878701298100
Deep analysis: Anthropic is building a secure OpenClaw. Four features in 30 days, each reverse-engineered from the open-source agent. Dispatch = OpenClaw phone control. Channels = OpenClaw messaging. Computer use = OpenClaw OS access. Plugin marketplace = OpenClaw skills. The strategy: let open source take the arrows, ship the enterprise-safe version. OpenClaw proved 250K devs want phone-to-computer agents. Also proved it produces RCEs and 20% malware in skill ecosystems.
@spect3ral [Claude Code]
https://x.com/spect3ral/status/2036508379963932910
Built 140 ad creatives in 11 minutes using Claude Code skills: ad copy matrix (50+ variations from one brief), creative templates (React components), bulk image generation, hook optimizer, platform formatter (Meta/Google/LinkedIn), creative scoring, and performance reporter. What took a full afternoon now takes a coffee break.
@oliviscusAI [Claude Code]
https://x.com/oliviscusAI/status/2036395922016686548
Dexter: open-source Claude Code for finance. Finds undervalued stocks, analyzes them in detail, builds investment thesis. The financial vertical for Claude Code is taking shape.
@axiaisacat [Claude Code]
https://x.com/axiaisacat/status/2036370665743376466
Claude-Code-Game-Studios: 48 AI agents, 37 workflows, full studio hierarchy with creative director, tech director, producer, gameplay programmer, level designer, narrative, audio, QA, and release. Supports Godot, Unity, and Unreal. 3.3K GitHub stars. One-person game studios are becoming real.
@steipete [OpenClaw]
https://x.com/steipete/status/2036268009854296257
Peter Steinberger corrected misinformation: OpenAI did not buy OpenClaw. OpenClaw is owned by the independent OpenClaw foundation, which he leads. Important clarification as the Claude vs OpenClaw narrative intensifies.
@mslipper [OpenClaw]
https://x.com/mslipper/status/2036491937176252474
Open-sourced iron-sensor: eBPF-based behavioral monitor for AI coding agents. Agents read SSH keys, write cron jobs, modify systemd units, escalate privileges. Most of the time this helps. Sometimes it installs a backdoor. iron-sensor records everything in the kernel. Stress tested with YoloClaw — 223 skills installed, 16K events, no malware found, and they can prove it.
@wickedguro [OpenClaw]
https://x.com/wickedguro/status/2036440441651147106
Postiz crossed $70K MRR — up from $20K just a month ago. Credits OpenClaw for changing his life. Concrete revenue proof from the OpenClaw ecosystem.
@heygeorgekal [Claude Code]
https://x.com/heygeorgekal/status/2036557957383512450
Building a 100% AI-developed ARPG: every model by Meshy, every line of code by Claude Code, Godot engine + MCP. Now working on souls-like combat with 3-hit combos, dodge rolls, stamina, hitstop.
@IceSolst [Claude Code]
https://x.com/IceSolst/status/2036448063988883517
Security alert: LiteLLM, used as an LLM proxy by corporations, was compromised. Everything passes through it — secrets, data. Imagine Claude Code via a compromised proxy inserting backdoors in every codebase developers are working on. Supply chain attacks on AI infra are the new frontier.
@witcheer [Claude Code]
https://x.com/witcheer/status/2036481005465338082
Hermes agent v0.4.0: expose your agent as an OpenAI-compatible API endpoint. Your agent becomes a model. Six new messaging adapters (Signal, DingTalk, SMS, Mattermost, Matrix, webhook) — 10 platforms total. Context compression rebuilt from scratch. /queue lets you stack prompts while agent is working.
@bozhou_ai [Claude Code]
https://x.com/bozhou_ai/status/2036349896752537801
Obsidian CEO kepano published official Agent Skills for Claude Code — first CEO to personally build skills following the specification. Five skills for Obsidian: markdown, bases, canvas, CLI, and defuddle (web-to-markdown). 16K GitHub stars.
🗣 User Voice
User Voice
1. Claude vs OpenClaw is the defining debate — Claude ships enterprise-safe features fast but stays closed. OpenClaw is open and flexible but has serious security gaps (20% malware in skills). Users want both worlds. (@AlexFinn, @aakashgupta)
2. Supply chain security is now critical — LiteLLM compromised, Karpathy exposed a Python package attack, iron-sensor created to monitor agent behavior in the kernel. The ecosystem is waking up to agent security. (@IceSolst, @mslipper)
3. Auto mode is the right direction but needs more nuance — users want permission intelligence, not binary approve-all or manual-approve. (@claudeai, @omarsar0)
4. Figma MCP is a paradigm shift for solo builders — backend devs designing production UI, one-person teams shipping full products. (@trq212, @FigmaJapan)
5. Auto-dream memory consolidation is the sleeper feature — periodic subagent that consolidates memory during idle time, mimicking human sleep. (@kr0der)
1. Claude vs OpenClaw is the defining debate — Claude ships enterprise-safe features fast but stays closed. OpenClaw is open and flexible but has serious security gaps (20% malware in skills). Users want both worlds. (@AlexFinn, @aakashgupta)
2. Supply chain security is now critical — LiteLLM compromised, Karpathy exposed a Python package attack, iron-sensor created to monitor agent behavior in the kernel. The ecosystem is waking up to agent security. (@IceSolst, @mslipper)
3. Auto mode is the right direction but needs more nuance — users want permission intelligence, not binary approve-all or manual-approve. (@claudeai, @omarsar0)
4. Figma MCP is a paradigm shift for solo builders — backend devs designing production UI, one-person teams shipping full products. (@trq212, @FigmaJapan)
5. Auto-dream memory consolidation is the sleeper feature — periodic subagent that consolidates memory during idle time, mimicking human sleep. (@kr0der)
📡 Eco Products Radar
Eco Products Radar
| Product | Mentions |
|---------|----------|
| OpenClaw | 200+ |
| Claude Code | 220+ |
| Figma MCP | 18 |
| Codex | 14 |
| Cursor | 11 |
| Binance AI Pro | 8 |
| Dexter (finance) | 6 |
| LiteLLM | 5 |
| Hermes Agent | 4 |
| iron-sensor | 3 |
| Obsidian Skills | 3 |
| Product | Mentions |
|---------|----------|
| OpenClaw | 200+ |
| Claude Code | 220+ |
| Figma MCP | 18 |
| Codex | 14 |
| Cursor | 11 |
| Binance AI Pro | 8 |
| Dexter (finance) | 6 |
| LiteLLM | 5 |
| Hermes Agent | 4 |
| iron-sensor | 3 |
| Obsidian Skills | 3 |
Comments