[JarryShaw]

Summary

Round 8 of the whole-repo prose sweep (#1103). Fanned out 8 shard agents by directory; most of the repo was clean (no changes). Where there were real findings:

• Picked up the round-7 carry-over: ipv6_ext.py/.rst's _lookup_next_layer target (only the link target flipped to the fully-qualified Protocol._lookup_next_layer; the display text, naming the concrete Internet subclass, was already correct and untouched), plus four more sites the docs: prefer Protocol over ProtocolBase in cross-references, per the owner's #1103 ruling #2420 ruling missed (corekit/sentinels.py, docs/source/ext.rst x3, schema/schema.py, schema/link/ethernet.py). Not exhaustive: transport.py's ProtocolBase._lookup_registry link (added by fix(reassembly): the deferred TCP analyser dispatches on AppType, not bare ports #2421) and roughly 11 more unqualified ProtocolBase.<member> roles in protocol.py itself are still outstanding for a future round.
• internet.py, link.py, transport.py, sctp.py, protocol.py: register()'s Arguments:/Returns: said Protocol subclass/instance while the signature types ProtocolBase and the adjacent Raises: clause already said so — fixed to match.
• protocol-layer-placement.rst: the Link − Application − ProtocolBase set difference is __proto_key__ (added by fix(foundation): check every register_protocol_code key before registering any (#2268) #2343), not empty as the page claimed.
• registry-protocol.rst: "every registry overrides _validate_value" contradicted its own next paragraph (the AppType transports inherit it instead).
• vendor/ipx/packet.py, vendor/reg/linktype.py: a self-contradicting claim and a stale counted-number in vendor docstrings.
• engines.rst, conf.py, util/changelog_md.py, CONTRIBUTING.md: a stale scapy version, a wrong file path, a false "matches the Sphinx we build with" claim, and a missing caveat about actionlint making Lint non-advisory.

No executable code changed — every edit is inside a docstring, comment, or .rst file. Verified via diff review, py_compile on every changed .py file, and a clean import pcapkit from this worktree's venv.

Behavioral defects found, deliberately NOT fixed here (reporting separately)

• 16 of 24 pcapkit/vendor crawlers' process(self, data) reads the module-global DATA instead of its own data argument — latent, harmless only because request() happens to return that same global.
• pcapkit/vendor/pcapng/filter_type.py: empty registry behind a live TODO.
• Three open "figure out how to handle this programmatically" TODOs in pcapkit/vendor/mh/{cga_sec,status_code,cga_extension}.py.
• Dead multiprocessing.Pool scaffolding in vendor/__main__.py (left disabled on purpose, but misleading next to the exit-code contract in its own docstring).

Test plan

• Searched for similar pull requests
• make pylint/make mypy/make isort — not run (prose-only change, no code touched)
• make test — not run (prose-only change; py_compile + import pcapkit confirm no syntax/behavior change)
• Changelog entry: N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• docs — documentation only

[github-actions]

Coverage: 94.22% (unit tier, Python 3.15, c075cf62a, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	19609	133	2620	120	98.84%
pcapkit/corekit	2605	74	818	37	96.52%
pcapkit/dumpkit	282	3	102	3	98.44%
pcapkit/foundation	3064	24	1068	26	98.79%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	17593	193	4678	179	98.32%
pcapkit/toolkit	924	36	334	10	95.55%
pcapkit/utilities	502	7	144	5	98.14%
pcapkit/vendor	4449	1961	1020	167	52.64%

Per-file detail: the coverage-html artifact of this run.

[JarryShaw]

Review: GOOD TO GO at c075cf62a. Diff itself is clean prose-only (AST guard: all 13 touched .py files structurally identical once docstrings are stripped; no Raises: clause touched, all 5 correctly still say ProtocolBase; the removed count in vendor/reg/linktype.py is genuinely removed, not just updated). Docs build byte-identical warning sets (default + nitpicky) vs main+PR merged. Full tests/project plus a 22-file text-reading sweep outside it: clean.

Fixed the one overclaim in the PR body myself (metadata-only, no re-review needed): "finished the round-7 carry-over" overstated it — transport.py's ProtocolBase._lookup_registry link and ~11 more unqualified ProtocolBase.<member> roles in protocol.py remain for a future round. Softened to say so explicitly.

Two corrections worth noting for next time rather than reopening this round: the claimed "tests/examples sites follow a documented CONTRIBUTING.md convention" doesn't exist as stated (89 role hits, 73 member references remain there, left alone on #2422's own "not rendered" reasoning instead) — and the changelog has 5 untouched ProtocolBase sites, not 4.