Skip to content

fix(abstract-eth): harden EVM TSS payment and consolidation checks - #10136

Merged
dpkjnr merged 3 commits into
masterfrom
CECHO-2460-evm-tss-verify-lean
Nov 10, 2026
Merged

dpkjnr merged 3 commits into
masterfrom
CECHO-2460-evm-tss-verify-lean

Conversation

@dpkjnr

@dpkjnr dpkjnr commented Nov 5, 2026

Copy link
Copy Markdown
Contributor

What

A narrower alternative to #10134 that hardens verifyTssTransaction only, assuming the caller passes correct txParams. Signing routing, sdk-core plumbing, the legacy EvmCoin verifier, FLR and statics are untouched.

Payments (transfer, and stored payment treated the same way)

  • Single native: destination and value match, and calldata is empty.
  • Single ERC-20: a canonical transfer(address,uint256) whose recipient and amount match, with zero native value, sent to the declared token contract (resolved from statics; a token coin declares itself).
  • Explicit calldata (dApp / WalletConnect): to, value and data must match exactly. Nonzero native value is allowed.
  • Native batch: sent to the network batcher, with batch(address[],uint256[]) arrays matching the intent and value equal to their sum.
  • Token batch: batchTransferFrom(address,address[],uint256[]) with matching arrays, zero value and the declared token contract. Top-level tokenName is not required.

Consolidation (verification.consolidationToBaseAddress): the existing base-address checks stay. An ERC-20 sweep must also be canonical, carry zero native value and call the declared token contract; a native sweep must carry no calldata. Sweep amounts are not compared.

Warnings only: a token that statics can't resolve, a network with no batcher configured, ERC-20 calldata where no token is declared (recipient, amount and value are still enforced), ERC-7984 payments, and intent types with no verifier (including contractCall / customTx). A warning never skips the other checks.

Deliberately not in this PR

  • Rejecting contractCall / customTx: existing upstream tests send customTx through this verifier and expect it to pass.
  • Legacy / type-0 signing-preimage routing, binding serialized bytes to signableHex on ECDSA, the EvmCoin legacy bypass, FLR atomic handling, and the Katana batcher address in statics.
  • Plumbing that separates the stored intent from caller params.

Test changes

  • New: sdk-coin-eth/test/unit/tssPaymentVerification.ts, which builds real hteth bytes and covers matches and definite mismatches for every rule above. It also checks that a known payment with the wrong function fails, that a missing batcher does not hide wrong batch amounts, and that explicit calldata may carry native value.
  • sdk-coin-xdc/test/unit/xdcToken.ts: the fixture used mainnet USDC's contract for txdc:tmt, so it now uses the token's contract from statics.
  • sdk-coin-bsc/test/unit/bsc.ts: a tampered WalletConnect payload is now rejected by the exact-calldata check, so the expected message changed.

Verification

  • Node 24.22.1 unit suites: abstract-eth 188, sdk-coin-eth 549 (33 new), sdk-coin-evm 40, sdk-coin-bsc 43, sdk-coin-xdc 65, sdk-coin-polygon 71, sdk-coin-flr 61, sdk-coin-arbeth 56, all passing.
  • No SDK replay against production data was run for this branch. Exact explicit-calldata matching relies on the earlier field-level SQL validation, which found no mismatches.

Ticket: CECHO-2460

🤖 Generated with Claude Code

verifyTssTransaction compared only the destination and amount of a
single native or ERC-20 transfer declared as `transfer`. It skipped
stored `payment` intents, batches, the token contract, native value on
token transfers, and the calldata shape, so mismatched bytes could reach
signing.

Payments: stored `payment` is treated as `transfer`. Explicit calldata
must match to/value/data exactly; a native send carries no calldata; an
ERC-20 transfer must be canonical, carry no native value and call the
declared token contract. Multi-recipient payments must use the network
batcher with matching batch/batchTransferFrom arrays, the summed value
for native and zero for tokens.

Consolidation keeps its base-address checks and adds zero native value,
canonical calldata and the declared token contract for ERC-20 sweeps,
and no calldata for native sweeps. Sweep amounts are not compared.

Unresolved token or batcher metadata and intent types without a
verifier log a warning and never suppress other checks. Signing routing
and legacy verification are unchanged.

Ticket: CECHO-2460
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@linear-code

linear-code Bot commented Nov 5, 2026

Copy link
Copy Markdown
Contributor

CECHO-2460

@github-actions

github-actions Bot commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

⚠️ Compatibility checks are failing on Node.js 27-nightly (experimental, non-blocking). This does not block merge, but it signals an incompatibility to fix before this release becomes supported.

View run

@claude claude Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

Tip: disable this comment in your organization's Code Review settings.

@dpkjnr

dpkjnr commented Nov 5, 2026

Copy link
Copy Markdown
Contributor Author

@claude review

@github-actions

github-actions Bot commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @dpkjnr's task in 29s —— View job


Review of #10136: harden EVM TSS payment and consolidation checks

I read the abstractEthLikeNewCoins.ts diff in full. I did not read the new test file, and I did not run the test suites. The approach is sound and the scope matches the PR description. The canonical-ABI re-encode check is a good way to catch trailing bytes, dirty padding and odd offsets. I found no blocking bugs, only the items below.

Should fix

  1. Consolidation warning prints "undefined". In the consolidation ERC-20 branch (abstractEthLikeNewCoins.ts ~L4030), resolveTssToken(txParams.tokenName) returns { isToken: false } with no reason when no token is declared and the coin is a base coin. The warning then reads undefined: the consolidation token contract was not verified. The single-payment and batch paths already fall back with token.reason ?? 'payment declares no token'. Use the same fallback here, for example token.reason ?? 'consolidation declares no token'.

  2. A consolidation with ERC-20 calldata passes without a contract check if no token is declared. The base-address and canonical checks still apply, but to can be any contract. This is the same trade-off you accept for payments. It is worth a test that pins the behavior (a warning, not a pass or a throw) so a later change doesn't alter it unnoticed.

Worth a look

  1. Unparseable amounts throw a raw error. In verifyTssBatchPayment (the reduce over recipients), BigInt(String(recipient.amount)) throws a bare SyntaxError for a malformed amount such as "1.5" or "". Every other mismatch goes through throwRecipientMismatch. tssValuesEqual already wraps this in try/catch, so a small helper that returns a mismatch on a parse failure would keep the error behavior consistent.

  2. A token batch with a native recipient is not rejected. In the token-batch path, a recipient with isToken: false just produces the "declares no token" warning. A mixed batch is therefore not flagged. If a batch moves exactly one asset, as the comment says, consider rejecting recipients that resolve to the native asset inside a batchTransferFrom.

  3. Warning noise. warnTssVerificationIncomplete now fires for every non-dedicated intent type, and for transfer with no recipients, on every verify call. That is fine for rollout visibility, but console.warn per transaction can be noisy in the express server. Consider routing it through whatever logger the module uses elsewhere.

Notes and nits

  • The explicit-calldata branch skips the token and contract checks and allows native value. The PR description states this. The safety rests on the caller's txParams being correct, and the PR defers stored-intent separation.
  • The BSC and XDC test edits make sense: the fixture used the wrong token contract, and the changed error message comes from the stricter check.
  • The PR says no replay against production data was run. For a change that now rejects transactions it used to accept (native consolidation with calldata, non-canonical ERC-20 encodings, payment now verified like transfer), a replay or a staged rollout would give more confidence than the SQL validation of explicit calldata alone.

Stored payment approvals reconstruct recipients without caller txParams.
Pass the signing coin's chain name in both ECDSA MPCv2 verification
branches so amount.symbol restores tokenName for confidential payments.

Cover both signing-byte routes with the real EVM verifier and preserve
existing caller-recipient precedence.

Ticket: CECHO-2460

Co-authored-by: Ralph CLI <ralph-cli@bitgo.com>
Ralph: 18.13.2-bitgo.6 openrouter/deepseek/deepseek-v4.1-flash-1017@default:high+smol:high+task:high openrouter/anthropic/claude-opus-5.5@slow:high+plan:xhigh openrouter/openai/gpt-6.1-terra@vision:low+advisor:medium openrouter/z-ai/glm-5.3-flash@commit:high+memory:low openrouter/inclusionai/ling-3.1-flash@tiny:low openrouter/moonshotai/kimi-k3.1-pro@designer:high openrouter/google/gemini-3-pro-image@image web/exa@web openrouter/microsoft/mai-transcribe-2@dictation openrouter/typesafe/jev-1.14@judge
@dpkjnr
dpkjnr requested review from a team as code owners November 5, 2026 18:49
@dpkjnr

dpkjnr commented Nov 5, 2026

Copy link
Copy Markdown
Contributor Author

@claude review again

@github-actions

github-actions Bot commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @dpkjnr's task in 36s —— View job


Re-review of #10136 (second commit d4235db)

I read the diff of HEAD~2..HEAD for abstract-eth and sdk-core. I did not run the test suites. I could not diff against origin/master because the checkout is shallow and has no merge base.

Resolved: ERC-7984 approvals from a base-coin wallet

The new chainName argument on resolveEffectiveTxParams in ecdsaMPCv2.ts (L994/L1004) fixes the issue raised in the earlier inline comment.

  • mpcUtils.ts:251 sets amount.symbol to the chain name for native sends and to the token name otherwise.
  • On a stored-request approval, a token recipient now gets tokenName from amount.symbol. A confidential token therefore reaches the token.confidential warning path and no longer fails with the "not a native send or ERC-20 transfer" error.
  • Native sends still produce no tokenName, because the symbol equals the chain.
  • The PR's three updated test stubs add getChain, which matches the code change.

Still open from the first review

  1. Consolidation warning can print "undefined". At about L4036, `${token.reason}: the consolidation token contract was not verified` is unchanged. resolveTssToken returns { isToken: false } with no reason when no token is declared. Use token.reason ?? 'consolidation declares no token', as the payment paths do.
  2. Raw SyntaxError on a malformed amount. In the native-batch branch of verifyTssBatchPayment, BigInt(String(recipient.amount)) inside the reduce throws a bare error for an amount like "1.5" or "". Every other mismatch goes through throwRecipientMismatch. Parse once up front and report a mismatch on failure.
  3. A token batch with a native recipient is only a warning. Take a batchTransferFrom where one recipient resolves to isToken: false. The code treats it as "declares no token" and warns. The comment says a batch moves exactly one asset, so that recipient should be rejected.

New observations

  1. MPCv1 ECDSA is not updated. ecdsa/ecdsa.ts:835 and :842 still call resolveEffectiveTxParams without chainName. If any EVM wallet signs through MPCv1, the same payment mapping can drop the token identity there. If MPCv1 is out of scope for EVM, a one-line note in the PR description would say so.
  2. Pin the ERC-7984 behavior with a test. Nothing in the diff covers a stored payment intent with a confidential token name that reaches verifyTssTransaction and produces a warning instead of a throw. That is the regression the earlier comment found, so a test belongs next to the chainName change. The same applies to a consolidation with ERC-20 calldata and no declared token, which should warn.
  3. Warning volume. console.warn fires on every verify of a non-dedicated intent type and on transfer with no recipients. ERC-7984 builds send empty recipients by design, so each of them warns. Consider the module's logger or a once-per-reason throttle.

Summary: The second commit fixes the blocking regression for MPCv2. The remaining items are minor (1–3) or test and coverage gaps (4–6). The earlier note stands: no replay against production data was run. Because this change rejects transactions that used to sign, a staged rollout would add confidence.
· branch CECHO-2460-evm-tss-verify-lean

@claude claude Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Findings marked 🟡 are optional suggestions and need no follow-up push.

Still open from earlier reviews (1):

If you have decided not to act on one of these findings, resolve its thread (a reply alone leaves it open) and the next review stops counting it. To review this commit again now, use Re-run on its "Claude Code Review" check.

Stored pending-approval signing reconstructs recipients from the persisted
intent because recreateTxRequest -> signTxRequest carries no caller txParams.
For a token payment the only surviving asset hint is amount.symbol, so
resolveEffectiveTxParams needs the signing chain name to recover tokenName.

The MPCv2 call sites already pass it, but the MPCv1 ECDSA callers did not.
With the stricter verifier on this branch, an ERC-7984 stored payment from an
MPCv1 wallet was rejected as "not a native send or an ERC-20 transfer", and an
ERC-20 payment lost its declared-contract check and warned instead. Pass
this.baseCoin.getChain() in both MPCv1 verification branches, matching MPCv2
and EdDSA, and widen the reconstruction JSDoc accordingly.

Ticket: CECHO-2460
@dpkjnr

dpkjnr commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

@claude review again

@claude claude Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

Still open from earlier reviews (2):

  • Unresolved: 2 minor or pre-existing.

@dpkjnr dpkjnr self-assigned this Nov 10, 2026

@Charan-Ratlan Charan-Ratlan left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

WP owned files look good

@dpkjnr
dpkjnr merged commit 1e8d7a8 into master Nov 10, 2026
29 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants