[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible — 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?

• fix — corrects a defect

---

Description of your pull request and other information

Fixes #2411.

#2366 (fixing #2297) put the house ESP: <lowercase description> prefix on ESP.make's ProtocolError messages and on SecurityAssociation.unavailable, but eight sites were outside its checklist. The same condition therefore read two ways depending on which direction it was hit from — make said ESP: no security association for SPI 0x..., read said no security association for SPI 0x....

The eight, and the text each now carries:

Site	Message
_read_trailer, plaintext too short	ESP: decrypted payload too short: SPI 0x00004321 decrypts to 1 octets, too short to hold a pad length and next header
_read_trailer, pad length beyond the plaintext	ESP: invalid pad length: the decrypted payload for SPI 0x00004321 declares 255 octets of padding but holds only 3; the encryption key is most likely wrong
_read_trailer, padding pattern (returned and warned with)	ESP: invalid padding: the decrypted payload for SPI 0x00004321 does not follow the monotonically increasing sequence of RFC 4303 s2.4
ESP.read, NO_SA	ESP: no security association for SPI 0x00004321
ESP.read, TRUNCATED	ESP: payload too short: 4 octets, shorter than the 16-octet ICV the security association declares
ESP.read, AUTH_FAILED (separate integrity)	ESP: integrity check value does not verify for SPI 0x00004321 sequence 7
ESP.read, AUTH_FAILED (AEAD tag)	ESP: authentication tag does not verify for SPI 0x00004321 sequence 3; the encryption key is most likely wrong
SecurityAssociation.__init__, integrity-key warning	ESP: invalid integrity key: AUTH_HMAC_SHA2_256_128 expects a 32-octet key, got 16 octets; the ICV will very likely fail to verify

_read_trailer is a staticmethod, so it spells the prefix literally, as every SecurityAssociation site does; ESP.read is an instance method and uses {self.alias}:, as ESP.make does.

The warning is the one that needed this most: warn reports the caller's frame rather than the protocol, so the message text is the only thing naming ESP at all.

Two messages were reworded rather than only prefixed, to avoid ESP: ESP payload ... and ESP: decrypted ESP payload ...; the three reading <thing> for SPI x does/declares y were turned into the description: detail house form. Every substring the existing tests match on is preserved.

Tests. tests/protocols/internet/test_esp_message_prefix_2411_unit.py, 13 tests, each message pinned whole rather than by substring — matching fragments is exactly why the omission survived #2366. Fail-without-fix confirmed: against unpatched esp.py, 10 failed and 3 passed, each failure a missing prefix; the 3 that passed are the controls asserting the guards did not widen (a correct ICV still verifies, a consistent trailer is still read, a right-sized key still does not warn). tests/protocols/ in full: 2033 passed / 51388 subtests / 0 failed.

All but one of the new tests runs without cryptography: an ENCR_NULL SA reaches the trailer and separate-integrity paths on hmac and hashlib alone. Only the AEAD-tag case is gated on HAS_CRYPTO.

[JarryShaw]

Review: NEEDS CHANGES at 144d27aa (Sonnet, read-only cross-review).

Real blocker: tests/test_tier_guard.py::DependencyGateFalsifiabilityTests::test_removing_crypto_from_the_test_job_is_caught fails — the new test file adds a second skipUnless(HAS_CRYPTO) gate, but that test's expected set hardcodes {test_esp_unit.py}. Reproduced locally (199 passed/1 failed on PR head, 200 passed on main) and confirmed in real CI — Python 3.11-3.14 legs all show exactly this one failure. The "0 failed" in the PR's own report only ran tests/protocols/, which skips this top-level test file. Fix: add the new module to the expected set in test_tier_guard.py, or move the AEAD-gated test into test_esp_unit.py instead of a separate file.

Everything else checks out: all eight message prefixes correct (literal ESP: vs {self.alias}), every rendered value (SPI, octet counts, pad length, sequence, key size) verified correct, all 5 existing assertIn substrings in test_esp_unit.py survive, fail-without-fix confirmed (10 failed/3 passed on revert), no regression vs main (identical pre-existing FileNotFoundError failures in fresh worktrees, resolved by generating samples), no scope creep beyond esp.py + the new test file.

Non-blocking nits: "too short" appears twice in the trailer message; the ICV/tag messages are prefixed but not restructured into the description: detail form the three trailer messages use.

[github-actions]

Coverage: 91.73% -- tests failure, so this is partial (unit tier, Python 3.15, 144d27aa4, Unit Tests run failure)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	19609	879	2620	866	92.13%
pcapkit/corekit	2596	74	818	37	96.51%
pcapkit/dumpkit	282	3	102	3	98.44%
pcapkit/foundation	3045	24	1062	26	98.78%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	17576	193	4668	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 d77427d5 (Sonnet, round 2 — narrow final check). Delta from round 1 is exactly tests/test_tier_guard.py (+8/-1); esp.py and the new test module are byte-identical. Blocker confirmed fixed — the file now passes 135/2497/0. Independently judged (not just trusted): extending the gate set is the right call per #2324's own design (a per-gate tuple set, several gated modules is expected), the new assertIn is non-vacuous, and none of the other 3 open PRs adds a conflicting crypto gate. Spot-checked both ESP test files (49 passed) and tests/protocols/ in full (2033 passed, 0 failed) — did not re-run the full 36-minute suite from round 1's own report, so that figure stands unverified but the narrower checks are sufficient here.