Bug: Script and sub-workflow templates fail in explicit context mode

[PolyphonyRequiem]

Bug Description

In explicit context mode, script agent args and sub-workflow input_mapping templates that reference {{ workflow.input.X }} fail with TemplateError because workflow.input is an empty dict {}, even though the inputs were correctly passed via --input.

Reproduction

workflow:
  name: explicit-script
  entry_point: detector
  context:
    mode: explicit
  input:
    work_item_id:
      type: number
      required: true
agents:
  - name: detector
    type: script
    command: pwsh
    args:
      - "-Command"
      - "Write-Output '{{ workflow.input.work_item_id }}'; exit 0"
    routes:
      - to: $end

conductor run explicit-script.yaml --input work_item_id=42

Expected: Script outputs 42
Actual: TemplateError: 'dict object' has no attribute 'work_item_id'

Root Cause

WorkflowContext.build_for_agent() in explicit mode initializes workflow.input as {} and only populates entries explicitly declared in the agent's input: list via _add_explicit_input(). Script agents and sub-workflow input_mapping templates typically don't declare workflow inputs in their input: list because they are local template renders (no LLM prompt/token cost concern).

The same {{ workflow.input.X }} template works fine in:

• accumulate mode (default) — workflow.input is always fully populated
• last_only mode — same
• LLM agent prompt fields in explicit mode — if workflow.input.X is declared in the agent's input: list

Impact

Any workflow using context.mode: explicit with:

• type: script agents whose args reference {{ workflow.input.X }}
• type: workflow agents whose input_mapping references {{ workflow.input.X }}

...will fail at runtime with a TemplateError.

Workaround: Add workflow.input.X to the script agent's input: list, or switch to accumulate mode.

Environment

• Conductor v0.1.9
• Python 3.14.2
• Windows 11

Fix

PR #119