Senpi task argument padding QA

What was tested

This evidence set covers GPT-style optional-field padding at the Senpi task tool boundary, including single spawn, batch spawn, genuine ambiguous input, and the task prompt guidance consumed by the model.

What was observed

The isolated branch starts from origin/dev at 915c5e4a9f922099bd15b96fdd4c0b128c70e034 with a clean status. RED, GREEN, manual harness output, and cleanup receipts are added as the scenarios run.

Why it is enough

The focused tests pin semantic normalization at the argument boundary. The manual QA drives the real Senpi task surface to prove that provider-shaped calls spawn children instead of returning the prompt/tasks XOR error.

What was omitted

No credentials, raw environment dumps, auth headers, or private model traffic are stored. Runtime captures are sanitized to the invocation shape, task ids, terminal statuses, and decisive error or success text.