feat: support priced ERC-20 and native escrows

[g4titanx]

Summary

The SDK now deploys EscrowERC20 for a single ERC-20 transfer, EscrowNative for a single native transfer, and EscrowBatch for multiple rows. The selected type is passed consistently to pricing and obfuscation, stored in TransferSecrets, and reused during compliance and resume. A quote returning a different artifact type is rejected before deployment.

[g4titanx]

@claude review

[claude]

Claude finished @g4titanx's task in 3m 36s —— View job

---

Claude finished reviewing this PR

• Gather context (diff, changed files)
• Understand the request (code review)
• Review core logic changes (transfer.ts, bond.ts, api.ts, types.ts, errors.ts)
• Review test coverage changes
• Post final review feedback

Note: npm run check/npm test could not be run — Bash tool calls beyond read-only git/grep operations require interactive approval that isn't available in this context. If you'd like me to verify the build/tests, update the --allowedTools config to permit npm run *.

---

Summary

Clean implementation of EscrowERC20/EscrowNative/EscrowBatch selection. selectEscrowType (src/transfer.ts:94-97), the bond.ts rename to deriveBlindedSigners, and the EscrowKind threading through api.ts/nomad.ts compliance calls are all consistent with the stated policy in CLAUDE.md (single ERC-20/native row vs. multi-row batch). The fresh-quote path correctly rejects a mismatched artifact from pricing (INVALID_PRICING_QUOTE, covered by a new test in test/transfer-pricing.test.ts). The mock API's per-type constructor encoding looks correct and mirrors the real contracts' expected ABI shapes.

Finding: resume validation doesn't cross-check escrowType: "batch" against row count

quoteFromResume (src/transfer.ts:207-237) builds escrowMatchesRows like this:

const escrowMatchesRows =
  resume.escrowType === "batch" ||
  (rows.length === 1 &&
    ((resume.escrowType === "native" && isNativeToken(rows[0].tokenAddress)) ||
      (resume.escrowType === "erc20" && !isNativeToken(rows[0].tokenAddress))));

For "native"/"erc20" this correctly requires rows.length === 1 and a matching token kind. But for "batch" it short-circuits to true unconditionally — there's no check that rows.length > 1, and no check that the resumed row count matches whatever rows.length was used to derive the original batch signers. A resume call with escrowType: "batch" passes validation regardless of how many rows are supplied this time (even a single row).

This isn't just theoretical — the test suite's own VALID_RESUME fixture (test/transfer-pricing.test.ts:14-29) sets escrowType: "batch" while being exercised against single-row params (tokenAddress/recipientAddress/amount, not transfers[]) throughout the "rejects resume data missing %s" parameterized tests, which only works because this branch is unguarded.

Consequences beyond the validation gap itself: buildContext uses escrowType = params.resume?.escrowType ?? selectEscrowType(rows) and skips selectEscrowType/row-count reconciliation entirely on resume, and context.rows.length (the row count as seen on this call) later feeds polling's total (src/transfer.ts:621). If a caller resumes a batch transfer with a different row count than was actually deployed (typo, stale params, partial transfers[]), nothing raises INVALID_RESUME — it can silently proceed into compliance/signal/poll with a row count that doesn't match the deployed EscrowBatch.

Suggested fix — require the row count to actually match for batch too, e.g. via a rows.length > 1 check for the batch branch (exact row-count parity can't be verified from TransferSecrets alone since it doesn't persist the original row count, but the rows.length > 1 guard at least matches the invariant selectEscrowType enforces on the fresh-quote path):

const escrowMatchesRows =
  (resume.escrowType === "batch" && rows.length > 1) ||
  (rows.length === 1 &&
    ((resume.escrowType === "native" && isNativeToken(rows[0].tokenAddress)) ||
      (resume.escrowType === "erc20" && !isNativeToken(rows[0].tokenAddress))));

Fix this →

Minor observations (no action needed)

• The bond.ts rename (deriveBatchBlindedSigners → deriveBlindedSigners) is applied consistently; no stale references remain in src/ or test/.
• INVALID_PRICING_QUOTE reuses the free-form MirageError code string pattern already used elsewhere (code: string, no enum), so it's consistent with existing error handling.
• Doc updates in CLAUDE.md/README.md accurately reflect the new escrow-selection policy.

Branch: feat/priced-single-escrows

[ozwaldorf]

ci failing

[g4titanx]

ci failing

would need to merge escrow pr first

[ozwaldorf]

ci failing