Release Mail Tools 0.2.0

public/docs/ARCHITECTURE.md

@@ -1,11 +1,21 @@
 # Architecture
 
 The React shell lazy-loads one workbench. `core/mime.ts` turns bounded source
-text into immutable-looking header and MIME-part records; parsing never creates
-DOM. Transfer decoding consumes a global byte budget. Diagnostics, comparison,
+bytes into immutable-looking header and MIME-part records; parsing never creates
+DOM. A one-byte internal syntax view preserves arbitrary octets, while every raw
+multipart entity records exact offsets into the retained source buffer.
+Transfer decoding consumes a global byte budget. Diagnostics, comparison,
 canonicalization and redaction are separate pure modules.
 
+`core/mbox.ts` splits bounded mbox/mboxrd containers and derives a presentation
+tree from Message-ID, In-Reply-To and References. `core/dkim.ts` implements
+bounded RFC 6376 canonicalization and uses WebCrypto for RSA-SHA256 or
+Ed25519-SHA256 verification. The key record is explicit user input; the app has
+no DNS client and makes no network request.
+
 Only the selected HTML text is passed through DOMPurify. It is then embedded in
 an iframe without sandbox capabilities and with an inner policy that denies all
-connections and active content. Attachment downloads are explicit Blob URLs
-with sanitized filenames. The service worker caches only same-origin app files.
+connections and active content. Explicit `cid:` image references can resolve to
+small safe image parts already present in the message; no external URI is
+retained. Attachment downloads are explicit Blob URLs with sanitized filenames.
+The service worker caches only same-origin app files.

public/docs/PRIVACY-SECURITY.md

@@ -5,9 +5,16 @@ memory. There is no storage, telemetry, analytics, remote-resource request or
 server integration. Closing/reloading the page clears the workbench.
 
 Limits reduce accidental resource exhaustion but do not make Mail Tools a
-forensic parser. HTML loses links, styling and embedded resources by design.
+forensic parser. HTML loses links, styling and external embedded resources by
+design; only bounded safe image parts referenced by `cid:` can be embedded.
 Downloaded attachments remain untrusted. Authentication results and Received
-headers are self-asserted text; no DNS or cryptographic verification occurs.
+headers are self-asserted text.
+
+The optional DKIM lab performs local cryptographic verification against a DNS
+TXT value pasted by the user. It never resolves DNS, so a passing result means
+only that the message bytes match that supplied key. It does not prove that the
+key is current, authoritative or obtained securely, and it does not turn other
+authentication headers into verified evidence.
 
 Focused redaction removes a fixed set of top-level transport/authentication
 headers. It does not rewrite nested `message/rfc822` content and can leave names,