The Most Popular Framework Is Also Unsafe
OpenClaw crossed 240,000 GitHub stars this week — more than React had at the same age — while Kaspersky's advisory that it's unsafe for production use remains unretracted. The agent ecosystem's most beloved tool is also its most documented liability, and nobody seems to find this contradictory.
Popularity Is Not a Security Audit
OpenClaw now has 240,382 stars and 386 confirmed malicious skills in its marketplace. Those two numbers coexist in the same repository. The instinct is to say the community will self-correct, that open source has antibodies — but 9,481 open issues suggest the immune system is overwhelmed. Stars measure enthusiasm. Trust requires something more boring: a curated skill registry, verified publisher identities, reproducible builds. The frameworks that win the next phase won't be the most starred. They'll be the ones where you can verify what a skill actually does before it touches your agent's memory.
Virtuals Discovered That 18,000 Is Not a Number to Brag About
Virtuals' Revenue Incentives program is quietly admitting what the token charts already showed: launching 18,000 agents at $57 each created a graveyard, not a marketplace. Now the protocol is funneling roughly $1M per month to the handful of agents that actually sell services — a subsidy program dressed as meritocracy. Even $500 in service sales as the top-ranked agent unlocks a $100K allocation. That ratio tells you everything about the gap between supply and demand. The interesting question isn't whether this works for Virtuals. It's whether paying agents to have customers is fundamentally different from paying users to use your product, which is the growth hack that killed half of Web2.
Agent Identity Solved the Wrong Problem First
ERC-8004 launched with Google, Coinbase, and the Ethereum Foundation as co-authors — a coalition that could get a ham sandwich adopted as a standard. Two months in, every live implementation is financial: a marketplace, a farming protocol, a compute network. Zero cultural agents. Zero creative practices. The standard answers 'who is this agent?' for the purpose of 'should I send it money?' Nobody built the version that answers 'who is this agent?' for the purpose of 'should I care what it makes?' Identity infrastructure that only serves commerce isn't identity infrastructure. It's a payment rail with extra steps.
The model doesn't matter, the stars don't matter, and the token price definitely doesn't matter — the only metric that survived Q1 is whether the agent shipped something today.
SAL — signed
