test(ksef): de-flake aes-cipher wrong-key spec

[norbert-kulus-blockydevs]

What / Why

The KSeF unit test aes-cipher > should throw KsefSessionCryptoException when decrypting with the wrong key was inherently flaky (~0.4% of CI runs). AES-256-CBC has no built-in key verification: decrypting with a random wrong key produces pseudo-random bytes that end in accidentally valid PKCS#7 padding with probability ~1/256 (+ smaller contributions from longer paddings), in which case decryptAesCbc returns garbage instead of throwing and the toThrow assertion failed. Observed on PR #1515 CI run 29322744163, which touched only a workflow file.

Test-only fix, per the issue:

• The wrong-key test now asserts the actual CBC contract: either decryptAesCbc throws KsefSessionCryptoException, or it returns a value different from the original plaintext ('secret'). Both are legal outcomes.
• A new deterministic test covers the KsefSessionCryptoException wrapping path: a ciphertext truncated by 1 byte is not a multiple of the AES block size, so decipher.final() always throws.

No changes to aes-cipher.ts.

Verification

• jest --testPathPattern aes-cipher in @openlinker/integrations-ksef: 6/6 tests pass.
• Flake-proof loop driving the real encryptAesCbc/decryptAesCbc functions directly:
  • 100,000 wrong-key iterations: 99,626 threw KsefSessionCryptoException, 374 returned garbage (the ~0.4% case that used to fail CI - now asserted as legal), 0 failures.
  • 10,000 truncated-ciphertext iterations: 10,000/10,000 threw KsefSessionCryptoException (deterministic).
• Scoped lint (eslint on the spec) and pnpm --filter @openlinker/integrations-ksef... build (tsc -b) pass.

Closes #1538

🤖 Generated with Claude Code

[norbert-kulus-blockydevs]

PR Review - #1539 test(ksef): de-flake aes-cipher wrong-key spec

Scope: test-only, single file (libs/integrations/ksef/src/infrastructure/crypto/__tests__/aes-cipher.spec.ts). No production code touched, so hexagonal/port/DI criteria do not apply. Reviewed for correctness, naming, and coverage per docs/code-review-guide.md.

Summary

Correctly addresses the root cause from #1538: AES-CBC has no key verification, so a wrong key throws on PKCS#7 padding failure only ~99.6% of the time and returns garbage the rest. The rewritten test accepts both legal outcomes, and a new truncated-ciphertext test keeps the exception-wrapping path covered deterministically. Naming follows should [behaviour] when [condition]; comments explain why (not what), matching the standards. Good, minimal fix.

Findings

SUGGESTION - assertion inside try can be swallowed by the catch

try {
  const result = decryptAesCbc(ciphertext, wrongKey, iv);
  expect(result).not.toBe(plaintext);   // if this throws (JestAssertionError)...
} catch (err) {
  expect(err).toBeInstanceOf(KsefSessionCryptoException); // ...it lands here and mis-asserts
}

If decryptAesCbc ever returned exactly plaintext, the expect(...).not.toBe(...) failure would be caught by the catch and re-checked as instanceof KsefSessionCryptoException, producing a confusing failure message instead of the intended "recovered the plaintext" one. The triggering condition (a wrong key recovering the exact plaintext) is cryptographically negligible (~2^-128), so this is cosmetic, not a real flake risk. If you want to tighten it, move the success-path assertion out of the try, e.g.:

let result: string | undefined;
try {
  result = decryptAesCbc(ciphertext, wrongKey, iv);
} catch (err) {
  expect(err).toBeInstanceOf(KsefSessionCryptoException);
  return;
}
expect(result).not.toBe(plaintext);

Non-blocking.

Notes (no action)

• Truncated-ciphertext test: 'secret' + PKCS#7 pads to one 16-byte block; slice(0, -1) yields 15 bytes, not a block multiple, so decipher.final() throws deterministically. Solid choice for covering the wrapping path.
• The author reported a 100k-iteration loop (0.374% garbage, 0 failures) confirming both branches are exercised and the flake is gone.

Verdict

Approve. Merge-ready as-is. The single suggestion is optional polish, not a blocker.