archetech / archon / 30060909400 / 1

90%

main: 90%

Pull #767

github

docs: correct five factual overclaims found in review

All five verified against source; @macterra was right on each.

- MCP redaction (4.2, 12.8): errorMessage() is applied only on the error
  path; successful results are serialized unchanged, and show_wallet /
  show_mnemonic return the decrypted wallet and recovery phrase behind
  reveal:true. Narrowed the claim to secret-redaction-in-errors and said
  plainly that reveal tools are the security boundary, replacing "without
  the wallet or passphrase ever leaving the machine".

- Resolution conformance (3.3, 5.5, 15): dropped "strictly conformant".
  representationNotSupported appears nowhere in either implementation —
  preferredDidContentType() falls back to LD-JSON — so I removed it from
  the error list, and documented both deviations: the triple is labeled
  with document representation media types rather than
  application/did-resolution, and the fallback means the error is
  unreachable. Described as the Universal Resolver result shape.

- Envelope diagram (8.9): anoncrypt defaults to XC20P, not
  A256CBC-HS512; that is the authcrypt default. Split the ciphers onto
  their branches per cipher/didcomm.ts:381 and noted A256GCM is
  selectable.

- Test oracle (8.9): didcomm-node is gone from the repo. Described the
  oracle as historical and the committed TS-envelope fixtures decrypted
  by the Python tests as the current regression mechanism.

- Relay metadata (8.9): the claim that the relay cannot learn the sender
  was wrong. Authcrypt puts the sender DID URL in the visible protected
  skid header, exposed by getEnvelopeInfo() without decryption, and
  /deliver authenticates the sender DID. The relay learns who is talking
  to whom, so it reconstructs a social graph, not just a receive pattern.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

3103 of 3680 branches covered (84.32%)

Branch coverage included in aggregate %.

12360 of 13212 relevant lines covered (93.55%)

420.4 hits per line