## P0: channel decryption broken on prod (`OperationError` in
`decryptECB`)
### Symptom
```
Uncaught (in promise) OperationError
at decryptECB (channel-decrypt.js:89)
at async Object.decrypt (channel-decrypt.js:181)
at async decryptCandidates (channels.js:568)
```
Channel message decryption fails for most ciphertext blocks in the
browser console on `analyzer.00id.net`.
### Root cause
The original `decryptECB()` simulated AES-128-ECB via Web Crypto AES-CBC
with a zero IV plus an appended dummy PKCS7 padding block (16 × `0x10`).
Web Crypto **always** validates PKCS7 padding on the decrypted output,
and after CBC-decrypting the dummy padding block it almost never
produces a valid PKCS7 sequence, so Chrome/Firefox throw
`OperationError`. There is no Web Crypto knob to disable that check —
and Web Crypto doesn't expose raw ECB at all.
This is a well-known dead end: every project that needs ECB in browsers
ends up with a small pure-JS AES core.
### Fix
- Vendor a minimal pure-JS **AES-128 ECB decrypt-only** core into
`public/vendor/aes-ecb.js`.
- **Source:** [aes-js](https://github.com/ricmoo/aes-js) by Richard
Moore — MIT License (cited in the header comment).
- **Trimmed to:** S-boxes, key expansion (FIPS-197 §5.2), inverse cipher
(FIPS-197 §5.3). No encrypt path. No other modes. No padding logic. ~150
lines.
- `decryptECB(key, ciphertext)` keeps the same API surface:
`Promise<Uint8Array | null>`. It now delegates to
`window.AES_ECB.decrypt(...)`.
- `verifyMAC` and `computeChannelHash` keep using Web Crypto
(HMAC-SHA256 / SHA-256 — no padding pathology).
- Wired `vendor/aes-ecb.js` into `public/index.html` immediately before
`channel-decrypt.js`.
### TDD
- **Red commit (`36f6882`)** — adds `test-channel-decrypt-ecb.js` pinned
to the **FIPS-197 Appendix C.1** AES-128 known-answer vector. Compiles,
runs, and fails on assertion (`OperationError`) against the existing
implementation.
- **Green commit (`bbbd2d1`)** — vendors the pure-JS AES core and
rewires `decryptECB`. Test now passes (7/7), including a multi-block
assertion that two identical ciphertext blocks decrypt to two identical
plaintext blocks (true ECB, no chaining).
- Existing `test-channel-decrypt-m345.js` still passes (24/24).
### Files changed
- `public/vendor/aes-ecb.js` — **new** (vendored AES-128 ECB decrypt,
MIT, ~150 LOC)
- `public/channel-decrypt.js` — `decryptECB()` rewritten to delegate to
vendor
- `public/index.html` — script tag added for `vendor/aes-ecb.js`
- `test-channel-decrypt-ecb.js` — **new** TDD test (FIPS-197 KAT +
multi-block + edge cases)
### Risk / scope
- Decrypt-only, client-side, no server changes, no schema changes, no
config changes (Config Documentation Rule N/A).
- ECB is a single 16-byte block per packet for MeshCore channel traffic,
so the perf delta vs Web Crypto is negligible (a single `decryptBlock`
is ~10 round transforms on 16 bytes).
- HTTP-context safe (no Web Crypto required for ECB anymore).
### Validation
- All 7 FIPS-197 KAT + multi-block tests pass.
- Existing channel-decrypt M3/M4/M5 tests still pass (24/24).
- `test-packet-filter.js` (62/62), `test-aging.js` (18/18) unaffected.
- `test-frontend-helpers.js` has a pre-existing failure on master
unrelated to this PR (verified by stashing the patch).
---------
Co-authored-by: openclaw-bot <bot@openclaw.local>
Kpa-clawbot