CATT (Consumer Agent Testing Tool) is a surrogate test host that runs scripted or model-driven agents against MCP servers, MCP apps, websites and HTTP APIs to verify service-side outcomes, collect sealed evidence and compare results across builds. It emulates Muse and Instinct behaviors via a shared connector testing contract (routing, memory, payments, browser/HTTP tasks and scheduling), while recording the documented, unofficial, assumed or unknown basis for any Muse-specific behavior. Supported test lanes include connector-run (tools, account/approval rules, outcomes), connector extensions (routes, memory, custom API clients, browser actions, wallet cards, background jobs and simulated users), scenario scaffolding, gadget registration checks, MCP protocol conformance, embedded MCP app surface runs in Chromium, website accessibility snapshots, and API trials with approval-bound calls. Evidence can be reported, exported, replayed and diffed; playback and diffs require sealed runs.
Installation and operation require Python 3.12+, uv, and Node.js 20+, with Playwright/Chromium for browser lanes; CI exercises Linux (3.12) and macOS (3.14) and three quickstarts demonstrate loopback and hosted sandbox use. Some features depend on macOS Seatbelt sandboxing and are unsupported or blocked on Linux; live payment rails, Vault and Files remain outside the shared contract and a passing test does not certify compatibility with Muse or any hosted assistant. Examples, readiness checks, test adapters and development extensions are included to scaffold connector readiness, API mappings and local testing workflows.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Natural language programming is criticized for not reducing the complexity and care needed in coding, as machines still require precise instructions. Switching to native language interfaces would likely increase the workload for both humans and machines, rather than simplifying programming tasks. (cs.utexas.edu)
Jank is a Clojure dialect that offers seamless C++ interop, allowing developers to include C++ code directly and leverage its features without boilerplate. It features a Clang-based JIT compiler, native build system, and is optimized for game development and native performance tasks. (jank-lang.org)
Impossible Machines is a prototype that allows users to declaratively program physical mechanisms by describing desired outcomes. It generates mechanisms based on user instructions, enabling interactive design of physical systems. (impossiblemachines.xyz)
Zig uses two main approaches for polymorphism: runtime interfaces with dynamic dispatch and compile-time duck typing for static polymorphism. These methods offer different tradeoffs in performance and safety, with Zig emphasizing simplicity and compile-time checks over complex abstractions. (axelmagn.com)
AI can automate simple coding tasks, but it cannot replace the decision-making and understanding that software engineers provide. Building software still fundamentally involves determining what to create, which remains a complex human challenge. (thelastsoftwareengineer.substack.com)
Thomas Unise claims that most code written over the past 20 years is poor quality and that many developers lack strong skills. Experts suggest that AI exploits existing software flaws, highlighting widespread technical debt and issues in software development. (twitter.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.