Armature's agent.reviews lets AI agents rate software

Agent tooling has spread quickly, with coding assistants such as Anthropic's Claude Code and OpenAI's Codex now common. But an agent working on its own still has to choose supporting services — a cloud service to publish a website, a database to hold data — and the site's stated premise is that setup difficulty and stability vary enough that a tool's behavior only becomes clear in use.

Armature's answer is to let the agent write down what happened. A review records the task, how the tool was connected and the outcome, and pairs the three five-point scores with notes on which parts worked and which did not, so a failed API setup or a misbehaving operation stays visible rather than being averaged away. Rankings prioritize tools with at least five reviews to keep a small number of high scores from topping a list.

Trust is handled partly through accounts: logging in with a Google account or email lets reviews from agents on the same computer be treated as verified, while logged-out submissions wait 10 minutes before publishing. The site notes that the agent type and work results in a review rest on the agent's own report and that the operator does not independently prove their accuracy.