Super User Daily: 2026-04-23
Quiet day on the Super User feed. Anthropic's alleged removal of Claude Code from new Pro plans dominated the chatter, with builders asking if their workflows just got priced out. The few real use cases that did surface were heavy: a cross-market trading skill that turns a single chat into AAPL/BTC/oil correlation analysis, and a college student running 39 AI agents on a Mac mini for $21 a month against a $16k outsourcing alternative. The OpenClaw-as-orchestrator pattern keeps showing up in concrete numbers, not just hype.
@amyincrypto [Claude Code/OpenClaw]
https://x.com/amyincrypto/status/2046588633701097914
She installed the Hubble Data Service Skill and stopped writing custom data plumbing for cross-market analysis. In one chat she pulled an AAPL monthly K-chart with intraday tick (anomaly volume on 03/20, low on 03/30, high on 04/15), ran a one-hour BTC/USDT structural replay with RSI 67.94 and MACD golden cross, then computed a Pearson 0.42 between USO and BTC to conclude oil weakness has no significant negative transmission to BTC. Skill works on Claude Code, OpenClaw, and Hermes. The point she's making: VibeCoding can write code but cannot deliver a unified data layer — a hand-tuned skill solves what 100 hours of agent code generation cannot.
@qkl2058 [OpenClaw]
https://x.com/qkl2058/status/2046503704418037980
Narrative report on a college student who spent his $2400 scholarship on an iPad and Mac mini, pulled 39 AI agent repos from GitHub, hit enter, and went to sleep. Woke up to the equivalent of four junior-developer-days of work — no bugs, no questions. The orchestration layer was OpenClaw, which let all 39 agents see the project structure simultaneously and pick what to touch next. First paying client landed in a month, second a week later. A traditional outsourcing shop would charge $16k/month for the same volume; his stack runs at $21/month. The kid is in class while the Mac mini ships invoices.
@hittheaverage [Claude Code]
https://x.com/hittheaverage/status/2046592767724548333
Field observation, not a brag: he watched a founder build an entire product in Claude Code without writing a single prompt by hand — just smashed Ctrl+K and clicked accept all the way to 100 users in two weeks. Then spent three months trying to figure out what was actually broken. The accept-everything-shipped-fast / pay-back-with-debugging arc is becoming its own genre. Worth bookmarking next time someone says vibe coding ships products in a weekend.
@MosheTov [Claude Code]
https://x.com/MosheTov/status/2046514894787281248
Disclosed a class of RCE bugs in StdioServerParameters that ~200 of 3,000 open-source repos importing MCP/FastMCP/LangChain inherited — including Cursor and Claude Code apps. His point isn't the bug, it's that doing 30+ disclosure processes is impractical and devs/AI writing code can't be expected to build a secure StdioServerParameters every time. Security tools flag Subprocess.run as command execution but don't flag StdioServerParameters, so scale-error is baked in. Frames it like SQL injection before parameterized queries — a structural fix is needed at the framework layer, not 200 user-error patches.
🗣 User Voice
User Voice
@Angaisb_ and @terabytespower both flagged that Anthropic seems to have pulled Claude Code from new Pro plans — "can't believe it" / "あれ、Pro プランで使えなくなったの！？". When the entry tier disappears, the migration off Claude Code accelerates whether the team intends it or not.
@yidabuilds (zh): "Claude code 和 codex 也没比上 openclaw 巅峰期啊" — even with current limits, the comparison anchor users carry is OpenClaw at its peak, not Claude Code at today's. Branding loss that shows up in offhand comments.
@MosheTov: framework-level security ownership is missing. Every wrapper around MCP currently re-implements the same insecure surface, and there's no SQL-injection-style fix.
@Angaisb_ and @terabytespower both flagged that Anthropic seems to have pulled Claude Code from new Pro plans — "can't believe it" / "あれ、Pro プランで使えなくなったの！？". When the entry tier disappears, the migration off Claude Code accelerates whether the team intends it or not.
@yidabuilds (zh): "Claude code 和 codex 也没比上 openclaw 巅峰期啊" — even with current limits, the comparison anchor users carry is OpenClaw at its peak, not Claude Code at today's. Branding loss that shows up in offhand comments.
@MosheTov: framework-level security ownership is missing. Every wrapper around MCP currently re-implements the same insecure surface, and there's no SQL-injection-style fix.
📡 Eco Products Radar
Eco Products Radar
Nothing crossed the 3+ mention threshold in today's thin feed. Tools that surfaced once each: Hubble Data Service Skill, OpenClaw, Claude Code, Codex, Hermes, ChatGPT Pro, Polymarket/PolyCop, MCP/FastMCP/LangChain.
Nothing crossed the 3+ mention threshold in today's thin feed. Tools that surfaced once each: Hubble Data Service Skill, OpenClaw, Claude Code, Codex, Hermes, ChatGPT Pro, Polymarket/PolyCop, MCP/FastMCP/LangChain.
Comments