fix(executor): preserve 2-byte 0x7702 EIP-7702 marker when packing initCode

[AngeloAyranji]

Problem

EIP-7702 userOps that carry a paymaster are accepted by eth_sendUserOperation and then silently deleted at bundle time ("failed on chain estimation. deleting..."), never mining. Self-funded 7702 ops and non-7702 paymaster ops are unaffected.

Root cause

packUserOp right-pads op.factory to 20 bytes unconditionally (packedUserOpUtils.ts), turning the EIP-7702 marker 0x7702 into 0x7702 + 18 zero bytes. Paymasters that commit to keccak256(userOp.initCode) — e.g. Pimlico's singleton (SingletonPaymasterV7.sol:356, inherited by V8) — signed over the canonical 2-byte form, so the re-packed op fails paymaster signature validation with FailedOp("AA34 signature error") inside validateBundle's eth_estimateGas, and the op is cancelled without any client-visible error.

The account signature is unaffected because the EntryPoint normalizes initCode in its own userOpHash for 7702 ops (UserOperationLib.sol:48) — which is why only the paymaster combination dies (always AA34, never AA24).

Evidence

• Prod log (bundler-base-sepolia, 2026-08-04 08:45:58 UTC): op 0x04d50f8c… deleted with FailedOp(0, "AA34 signature error") from the bundle estimation.
• Byte-level A/B on Base Sepolia via Alchemy, same op, same node, same second: handleOps estimation with 2-byte initCode passes (0x615b3 gas); identical op with the 20-byte padded initCode reverts AA34 signature error.
• On-prod behavior matrix: 7702 self-funded → mined; 7702 + paymaster → silently dropped; paymaster from already-delegated EOA → mined.

Fix

Guard the pad with the exported INITCODE_EIP7702_MARKER constant (case-folded). Real 20-byte factory addresses take the identical path as before (the pad was a no-op for them).

Tests

New packedUserOpUtils.test.ts (5 cases): marker stays 2 bytes, odd-cased marker, marker + factoryData concatenation, real factory address unchanged, no factory → 0x. The 7702 cases were written first and observed failing with the padded output before the fix (TDD). tsc build clean; the executor suite's 8 pre-existing failures (anvil-dependent fixtures) are identical on clean develop.

Note

Upstream etherspot/skandha master has the same unconditional padding (hexRightPad(op.factory ?? "0x", 20)), so this is worth an upstream issue/PR as well.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SDWq76gdRTBEsNF6u6yJq7

[justalab-agents]

This PR fixes a bug where the EIP-7702 0x7702 marker in a UserOperation's factory field was being right-padded to 20 bytes during packing, invalidating paymaster signatures that commit to keccak256(initCode) and causing bundle-time AA34 reverts. The fix preserves the 2-byte marker as-is and adds unit tests covering normal factory addresses, the EIP-7702 case, and edge cases.

One thing the inline comments don't make obvious: the corresponding unpackUserOp function has a symmetrical gap — it still blindly slices 20 bytes for the factory address, which will misparse (or error on) any packed op using the 2-byte EIP-7702 marker, breaking the pack→unpack round-trip.

✅ No blocking issues found by the automated review — this is not a substitute for human review.