[JarryShaw]

Summary

Part of the #1103 whole-repo prose sweep. Fixes #2409: docs/source/changelog/1.5.0.rst had no entry for most PRs merged in the #2312-#2407 range.

40 added bullets citing 43 PRs, each cited exactly once: #2332, #2335, #2338, #2339, #2343-#2347, #2349-#2351, #2357-#2359, #2363-#2375, #2377, #2382, #2383, #2392, #2393, #2395-#2401, #2403, #2405, #2407.

5 already had entries, untouched: #2334, #2336, #2337, #2340, #2348.

12 skipped as not user-visible: version bumps #2387-#2391; prose-only #2333, #2376, #2379, #2380, #2381; test-only #2402; unreachable-branch removal #2378.

Three pairs share one bullet, each a direct continuation of its sibling:

• fix(protocols): resolve a str enum name when its argument defaults to None (#2356) #2393/fix(hip): resolve a str enum name in list-argument makers too #2401 — str enumeration name resolution where the namespace couldn't be inferred (pcapkit.protocols, Fixed)
• fix(protocols): makers name the rejected argument and value in error messages #2405/fix(protocols): four more maker guards name the rejected argument #2407 — a maker naming the rejected argument and value (pcapkit.protocols, Fixed)
• refactor(foundation): declare each registry's code kind as a class attribute, not a module table (#2343) #2392/refactor(foundation): declare _CODE_DESTINATIONS as a dunder on the enum registries themselves #2403 — the __proto_key__/__proto_dest__ dunders pairing a __proto__ registry with its enum (pcapkit.foundation, Added — the bare pre-change error text never shipped in a release, so this isn't framed as a behavior fix)

Dropped as factually wrong, found by two rounds of cross-review: #2370's "still fails" caveat (bullet kept, caveat removed — #2369 had already fixed it before #2370 merged); #2381's "left out of the rendered tables" claim (entry removed — conf.py already excludes the affected marker from autodoc, so nothing changed in rendered output); the original #2392 framing as a registrar-message behavior fix (superseded by the Added bullet above).

Not in this PR: #2404, closed unmerged in favor of #2405 (same issue, #2296); #2384-#2386, closed in favor of #2391; issues #2319-#2328 and #2341, closed by direct pushes to main (test/docs/chore only, no docs/source/changelog/ entry needed).

Test plan

• util/changelog_md.py --check passes; CHANGELOG.md regenerated and in step
• docs/source/changelog/1.5.0.rst diff vs main: 273 insertions, 0 deletions — purely additive
• tests/project/ in full: 2107 passed, 80 skipped, 10265 subtests, 0 failures
• A real Sphinx render (pinned docs venv) gives 0 warnings and the same nitpicky-reference count as main

[JarryShaw]

Review: NEEDS CHANGES at 3e8dd5af (Sonnet, read-only cross-review).

Verified good: census of all 60 merged PRs in range checks out (5 pre-covered + 43 new + 12 skipped, exactly as claimed); 1.5.0.rst diff is insert-only at bullet boundaries; CHANGELOG.md regenerates byte-identical; tests/project/ 2107 passed/80 skipped/10265 subtests; a real Sphinx render (pinned venv) gives 0 warnings and the same nitpicky-reference count as main.

Six entries have factual errors, each independently re-verified against the diff/HEAD:

• chore(vendor): HIP Group and Packet crawlers write #: in _missing_, not # (#2331) #2381 — "left it out of the rendered tables" is false; conf.py already excludes _missing_. Reword or drop.
• fix(ftp): place FTP_DATA in the application layer, in its own module #2357 — "shipped from 1.4.0" is wrong; FTP_DATA is in ftp.py since v1.0.0.
• fix(pcapng): handle make(block=<bytes>) for a DSB (#2354) #2370 — "SHB with no section context still fails" is false at HEAD; fix(pcapng): resolve byte order for a raw bytes option without self._byte set (#2353) #2369 already closed it. Delete the entry.
• fix(ipv6_ext): give the version gate's message the house form #2397 — add :issue:\2302`, re-sort after #2398; "named only the version it got" is wrong (old text already included only valid for IPv6, got version=N`).
• fix(toolkit): lead every adapter message with its engine's own __engine_name__ #2399 — "each adapter's messages lead with its engine's name" fails for dpkt and scapy.
• fix(ftp): reject a command that is not str or bytes with ProtocolError (#2290) #2365 — "everything that reached its fallback raised AttributeError" is false; the fallback was cmmd.value.

Plus #2392/#2403 (deviation 1): the message change is real, but the bare pre-change text never shipped in a release (landed in #2193, after the last tag), so the sentence framing it as a user-visible fix is wrong; also #2403 (skipped as internal) actually introduced __proto_dest__/Vendor.PROTO_DEST, which no entry currently credits. Drop the sentence or the entry, your call which.

Lower-confidence (subagent-reported, not independently re-run by me, but flagging): #2338 ("reassembled whole" — the datagram was never reassembled), #2398 (understates the message change), #2400 (omits changed default octet values), #2335/#2373/#2374 (overstate).

[JarryShaw]

Review: NEEDS CHANGES at 7f59b3e9 (Sonnet, round 2). All seven round-1 fixes verified correct, including both judgment calls — the kept #2370 bullet (a bisect confirms the removed sentence was false when it merged, 13 minutes after #2369 landed), and the #2392/#2403 shared "Added" bullet (the pairing is defensible: #2403's own commit message says "Mirrors __proto_key__", same framing as the #2393/#2401 precedent). 1.5.0.rst diff is 273/0, changelog_md.py --check passes, tests/project/ 2107 passed/80 skipped/10265 subtests.

Small fixes needed:

1. The new Added bullet says "the two declarations" and "the seven classes" — drop these per the 2026-10-15 no-counted-numbers ruling (policy, not accuracy — "seven" is currently true).
2. PR body is stale against head: lists chore(vendor): HIP Group and Packet crawlers write #: in _missing_, not # (#2331) #2381 as covered (it was deleted, correctly) but omits refactor(foundation): declare _CODE_DESTINATIONS as a dunder on the enum registries themselves #2403; still frames refactor(foundation): declare each registry's code kind as a class attribute, not a module table (#2343) #2392 as the registrar-message behavior entry; says "two pairs" share a bullet when there are now three; cites 263 0 instead of the real 273 0.

Optional nit: the #2399 bullet's "reassembly refusals" for pcap-ct/PyPCAP omits that their tcp_traceflow refusals were prefixed too.

[JarryShaw]

Review: NEEDS CHANGES → fixed at 813e9f89 (Sonnet, round 3). Content fully confirmed (no counted numbers remain, class-list enumerations verified against actual code, changelog_md.py --check and docutils parse clean, 273/0 insertions, 6 project tests 123/123 passed) — the only issue was the PR body: the "12 skipped" sentence named 11, missing #2381; and "out-of-range code citing #2404" had no referent. Both fixed directly in the body (also switched "closing #2409" to "Fixes #2409", which is the actual GitHub auto-close keyword — the original wasn't one).