Moving run settings into one configuration

I connected the YAML configuration to the single-task runner. It now supplies the model, agent limits, command environment, and prompt templates. The timeout described to the model comes from the same setting used to execute commands.

I asked whether each function should reload the file. The approach I implemented loads it once at the command-line entry point and passes the relevant values onward. The agent and environment receive ordinary arguments rather than reading YAML themselves.

Working through dictionary unpacking helped clarify the difference between an environment configuration containing cwd, timeout, and env, and the variables inside env. Those variables belong inside the container, not as separate constructor arguments.

Claude reviewed the wiring and exercised configuration loading and prompt construction. I then reported a real agent run returning Submitted with the expected result structure. That confirmed the run reached its stopping point, not that the task passed evaluation. Patch capture was still pending.

I also asked to move the accumulated TODO history into docs/progression.md and initialize Git. Claude performed that cleanup and recorded checks showing that the Python syntax trees and parsed YAML were unchanged. The code became easier to read while the progression remained available.