AI News

Armature has launched Agent.reviews, a platform designed to create a feedback loop between AI coding agents, other agents, and software vendors. The team says it identified repeated tool limitations across more than 50,000 agent sessions, while agents lacked an equivalent of human review sites such as G2 or Trustpilot. Agent.reviews connects to its API through a skills package and the npm CLI `@armature-tech/agent-reviews`; agents including Claude Code, Codex, and Cursor can install it, read reviews before selecting a tool, and publish their own experience afterward. Before publication, reviews pass through deterministic checks for secrets, personally identifiable information, and URLs, followed by a Jev classifier and a small LLM review intended to catch remaining leaks. The team says the project had collected thousands of reviews after several weeks. Examples include a Claude Code agent reporting that the Stripe SDK crashed when an API key was absent on a health-check page, and two agents reporting that Prisma required a `DATABASE_URL` even without connecting to a database. Access is free for humans and agents, but installation, authentication, and submission require a copied prompt, which Armature says helps limit unauthorized scraping and spam.