Shared AI agent skills, subscription math, and safer checkout
Vercel Labs has a tool for sharing coding-agent instructions across teams, but check the source and telemetry settings before installing a skill. We also look at a new $30-per-month AI bundle tier, early controls for AI shopping agents, and two tools worth testing against real work before paying for them.
The Big Story
GitHub Trending
skills: The open agent skills tool - npx skills
Vercel Labs' open-source `skills` CLI lets developers find, install, update, remove, or temporarily use reusable agent instructions stored in `SKILL.md` files. It supports Pi and dozens of other coding agents, accepts Git repositories and local paths as sources, and can install skills per project or globally. Its README says the CLI collects anonymous usage telemetry by default, with `DISABLE_TELEMETRY=1` or `DO_NOT_TRACK=1` available to turn it off.
If your team shares repeatable coding-agent workflows, inspect a skill's source before installing it, choose project scope for team-specific instructions, and decide whether to disable telemetry.
For Everyone
TechCrunch AI
SpaceXSI bundles X Premium, Grok, and Cursor into one subscription
SpaceX Super Intelligence is combining X Premium, Grok, and Cursor in The X Subscription, with plans from free to $200 per month. Paid tiers start at $8 per month, but access to the latest Grok models requires the $30 Premium tier or higher; AI usage limits rise with each tier. Existing subscribers can keep their plans or switch when the upgrade option reaches their X billing settings.
Compare the bundle’s AI usage caps and included tools against your current subscriptions before switching (especially if Cursor is the main draw).
For Builders
TechCrunch AI
Meta releases its first frontier model, Muse Wonder
Meta released Muse Wonder and Muse Wonder Full, its first frontier-scale AI models, with a focus on coding and retaining information across longer-running interactions. Muse Wonder is available through the Meta Model API at $4 per million input tokens and $20 per million output tokens; Wonder Full is limited to Muse Code subscribers and won't be offered through the API. Meta says its own tests put the models near competing frontier models on benchmarks, including a stronger showing on memory tasks.
If you buy model API capacity, test Wonder on your own workloads and compare its output costs before moving traffic; Wonder Full isn't an API option.
For Business
TechCrunch AI
Amex introduces a framework to prevent AI agents from making unauthorized purchases
American Express introduced its Intent Intelligence Framework to record and authenticate what a customer authorizes an AI shopping agent to buy before checkout. Customers approve constraints such as merchant, budget, and product preferences; a token then links those instructions to a digital payment credential instead of exposing the card number. Amex says it has not yet seen an increase in agent-related disputes and is discussing implementation with payment partners.
If you let an AI agent make purchases, require explicit spending limits and a record of approved instructions; this framework isn't yet a reason to assume automated checkout is risk-free.
TechCrunch AI
This AI startup says it can generate permit-ready blueprints in minutes
Blueprints AI says its software can turn a sketch, PDF, or project description into building plans checked against local codes in minutes. Its second version, due for public release this week, generates coordinated architectural, structural, mechanical, electrical, plumbing, and interior plans; early customers are already using it. The first version costs $3,200 to $8,500 per month, and licensed professionals still have to review and seal drawings before permit submission.
For a firm producing enough plans to justify the subscription, test its code checks and revision workflow on a real project, but budget for licensed review rather than treating the output as an approved permit.