test: JS/Python transcript parity across the language boundary

The handshake proof and admin signature transcripts are built independently in
crypto.js and in meshbay_common, and compared by producing identical bytes.
Nothing on the wire carries the transcript — that is the design — but it means a
one-byte disagreement between the two implementations is invisible to every
other test while causing a total outage: no browser could complete a handshake
with any node, and every file deletion would be rejected.

Nothing else in the suite crosses this boundary. The 278 Python tests would all
still pass.

Drives the real crypto.js under node (stubbing window and crypto, which the
module body touches but these functions do not) and compares against the real
Python for the same vectors: both roles, short and empty group ids, non-ASCII
group names and filenames — TextEncoder and str.encode must agree on UTF-8 —
and field splits that would collide under naive concatenation.

Verified to actually catch a mismatch rather than trusted for passing: removing
one length prefix from the JS fails 6 vectors, and changing a single byte of the
domain-separation prefix fails 6. crypto.js restored byte-identical afterwards.

Skips when node is absent, which is a coverage gap rather than a pass — worth
making a hard failure in CI (18.4).

Tests: 168 hub+common.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

0 files changed, 0 insertions, 0 deletions