proxybrowser

proxybrowser is a read-only forensic viewer over the MITM web-traffic capture set — the non-independence sensor that content-addresses every fetch made by cairns and λ agentic researchers. It renders research provenance: from a cairn source or a researcher run, link back to the actual web-traffic logs (which bytes, from which URLs, when, by which bucket), shown redacted.

It is not a WARC replay engine and not a new provenance store — it reads the existing capture-log.sqlite, a cdx.sqlite seek-index, and the WARCs.

Every openrouter.ai research fetch carries a live Authorization: Bearer sk-or-… and set-cookie. The render path routes through a redaction gate: denylisted headers collapse to a stable fingerprint (so "same key across fetches" stays visible without exposing the secret), and token shapes are scrubbed from bodies. There is no raw route.

byte-identical — a fetch's content-address equals a cairn source's excerpt_sha256. Rare today (cairn hashes a curated excerpt, the proxy the full document).

bucket / url+time — high-recall: the bucket name is the research context; the located WARC record is matched by url + nearest timestamp.