Built with AI · Ready for AI

The AI story

This entire suite was designed, ported and tested with AI assistance — and it's built to be easy for AI coding agents to consume.

How it was built

Every library here was ported from its Node.js original with Claude (Anthropic) driving the implementation: reading the reference behaviour, writing idiomatic Go, generating table-driven tests, and verifying wire-level compatibility against the real Node clients.

Faithful behaviour

Ports were checked against the original semantics — not just "looks right".

Verified, not vibes

Every package ships tests. CI runs build + vet + test on Go 1.23 and 1.24.

Self-documenting

A go/doc generator produces the API sites, so docs never drift from the code.

The site assistant

The /ask assistant answers questions about the libraries, grounded in a live search over the symbol corpus, and cites in-site deep links. On deployments where it is switched on, it can also run code to verify it: it writes a small Go, Node/TypeScript or Python program and executes it in an isolated, ephemeral sandbox, then reports the real stdout/stderr and exit code before presenting the snippet.

It can also read the project's recorded security findings: ask whether a library has a known issue, whether a particular version is affected, or whether something has been fixed, and the answer comes from the parity security manifests behind the Security page — with the affected version range quoted exactly and the fixed/unfixed status derived from the library's released version, never guessed. It will say plainly when the latest release is still affected, and it does not speculate past what a finding records.

Code execution is opt-in. It runs only when the deployment sets the CHAT_SANDBOX_ENABLED flag; otherwise the assistant answers normally and simply says it can't run code. The sandbox is ephemeral, receives none of the site's secrets, and is bounded by hard timeouts, output-size caps, and a per-conversation run limit.
Transparency: AI wrote the code, humans set the direction, and the test suites + protocol interop checks are what make it trustworthy. Treat these as you would any dependency — read the docs, run the tests, pin your versions.
AI