Consider routing access-control writes through shared runCapabilityWrite (typed-error mapping)

Context

Note: The code referenced below is not yet committed or pushed to remote. It currently lives on the local feature branch 002-ri-evm-capabilities (RI EVM capabilities POC) and has only been committed up to the IRS capability (692f1a6); the ERC-3643 work that introduces runCapabilityWrite is still uncommitted at the time of filing. Paths/symbols below will not resolve on main until that branch lands. This issue exists to capture the follow-up so it isn't lost.

While implementing the RI capabilities (IRS, ERC-3643), we extracted a shared write skeleton:

• packages/adapter-evm-core/src/shared/executor.ts → runCapabilityWrite(params, mapError)

It centralizes the common write control flow: submit assembled calldata via the injected executor → catch → log → map the failure to a typed error. Both EvmIRSService.execute and EvmErc3643Service.execute now use it, each supplying its own capability-specific mapError policy (IRS → IdentityOperationFailed; ERC-3643 → RecipientNotVerified / HolderFrozen / InsufficientBalance / ComplianceModuleRejected / fallback RICapabilityOperationFailed).

Problem / Opportunity

The third write-capable service, EvmAccessControlService, does not use this skeleton. Its write entry point delegates straight to the executor with no catch and no error mapping:

// packages/adapter-evm-core/src/access-control/service.ts
protected async executeAction(txData, executionConfig, onStatusChange?, runtimeApiKey?): Promise<OperationResult> {
  return this.executeTransaction(txData, executionConfig, onStatusChange, runtimeApiKey);
}

It was intentionally left as-is during the RI work because adopting runCapabilityWrite there is a behavioral change, not a like-for-like refactor: it would start catching every write error, logging it, and wrapping it into a typed error. That requires its own decisions and is out of scope for the POC branch.

Proposed follow-up

Evaluate giving access-control writes consistent typed-error mapping via runCapabilityWrite:

• Define an access-control mapError policy (likely mapping known reverts → existing AccessControlError subclasses such as OperationFailed, with a sensible fallback).
• Route executeAction through runCapabilityWrite with that mapper.
• Update/extend access-control write tests that currently assert on raw/unwrapped errors.
• Confirm no consumer (e.g. role-manager, ui-builder) depends on the current raw-error shape from access-control writes.

Acceptance criteria

• Access-control write failures surface as typed AccessControlErrors consistent with the IRS/ERC-3643 pattern, or a deliberate decision is recorded that access-control should keep propagating raw errors (with rationale).
• All existing access-control tests pass (updated as needed).
• No behavioral regression for downstream consumers.

Start by reading packages/adapter-evm-core/src/shared/executor.ts and packages/adapter-evm-core/src/access-control/service.ts, then inspect the existing access-control write tests and downstream consumers such as role-manager and ui-builder. Determine whether access-control should adopt typed error mapping or deliberately retain raw errors, and verify the chosen behavior with passing tests and no consumer regressions.