Workflow inputs not resolved - modelIdOrName receives literal expression string

Description

When running a workflow that uses ${{ inputs.<name> }} expressions in task.modelIdOrName fields, the expressions are not resolved. Instead, the literal string ${{ inputs.deviceModel }} is passed as the model name, causing "Model not found" errors.

Steps to Reproduce

1. Install the @john/tailscale extension: swamp extension pull @john/tailscale
2. Create the required models (ts-device, ts-user, ts-dns, ts-acl, ts-settings)
3. Run the tailnet-overview workflow with inputs:

swamp workflow run tailnet-overview --input '{"deviceModel":"ts-device","userModel":"ts-user","dnsModel":"ts-dns","aclModel":"ts-acl","settingsModel":"ts-settings"}' --json

Expected Behavior

The ${{ inputs.deviceModel }} expression in the workflow YAML should resolve to ts-device, and the workflow should execute the list method on the ts-device model.

Actual Behavior

Every step fails with errors like:

"error": "Model not found: ${{ inputs.deviceModel }}"

The expression is passed through literally without evaluation.

Workflow Definition

The tailnet-overview.yaml workflow uses this pattern:

steps:
  - name: list-devices
    task:
      type: model_method
      modelIdOrName: ${{ inputs.deviceModel }}
      methodName: list

Workaround

Running each model method individually works fine:

swamp model method run ts-device list --json

Environment

• swamp version: 20260227.033008.0-sha.c637ddd9
• OS: macOS Darwin 25.3.0

Summary

This bug affects the workflow execution engine's input expression resolution. The fix would involve ensuring that ${{ inputs.<name> }} expressions in workflow step task fields (specifically modelIdOrName) are evaluated before the step attempts to resolve the model.