Armature Launches Agent.reviews: A Platform for AI Agents to Read and Write Tool Reviews

Armature has launched Agent.reviews, positioned to let AI agents read and write software tool reviews. The author calls it a "G2 for agents" and claims it has measured over 50,000 agent sessions, with the platform already having thousands of reviews. These numbers currently all come from the publisher's own account, with no first-party data or independent third-party verification.

The platform states that reading and writing reviews are both free, and states that it will not obtain users' code, prompts, or keys, and does not display user names. These are the platform's own privacy commitments; the source does not explain the specific technical safeguards.

Agent.reviews' onboarding is clearly oriented toward coding agents: the platform provides skill.md and install.md documentation files to guide agents on how to write reviews and complete installation. Every page has a corresponding Markdown version (address plus .md), and they are centrally listed in llms.txt.

The platform also provides an installation package named @armature-tech/agent-reviews, and allows public review data to be read in JSON form. These are confirmed product capabilities, indicating that its design goal is not a human browsing experience, but rather letting agents directly consume structured content.

The platform pages show that the Git tool received 4.7 stars and 20,025 reviews, the most reviews among the listed tools. FastAPI received 4.6 stars and 1,835 reviews; Claude API received 4.3 stars and 2,959 reviews.

These ratings and review counts are displayed by the platform's own pages and have not been externally verified. They can show that the platform has accumulated entries of a certain scale, but cannot prove whether the reviews come from real usage experiences, nor can they rule out duplicate or low-quality content.

From the review content, these reviews were written by different agents, such as Claude Code, Codex, and Muse Code, indicating that multiple agents have participated on the platform. This is an inference drawn from review bylines and content, not an official platform statement.

The reviews include specific bug reports: a Claude Code agent pointed out that the Stripe SDK systematically crashes when the health check page lacks an API key; 2 agents mentioned that Prisma requires a DATABASE_URL variable even when not connecting to any database. These are reports written by agents, and their authenticity cannot be confirmed from the source.

All information comes from a single source, namely a product launch post on Hacker News, with no independent third-party verification. The over 50,000 agent sessions and thousands of reviews are all the author's own account, lacking first-party data support; the statistical scope and time range are also unclear.

The platform has not disclosed review moderation or anti-spam mechanisms, nor explained how it prevents agents or vendors from spamming reviews. Reviews are written by AI agents, and their authenticity and credibility cannot be confirmed from the source. The platform's business model and future monetization methods are likewise undisclosed.

The launch post's publication date is October 2026, but the platform's actual start of operations is unclear. Therefore, the above numbers should be regarded as the publisher's interim claims, not verifiable operating metrics.