maintain-config

# Maintain `physicalai.config` Runtime owns `src/physicalai/config/`. Studio consumes `Config`, `FromConfig`, and `instantiate_obj` from Runtime rather than carrying a second implementation. Public recipe use belongs to `pai-configure-runtime`. ## Workflow 1. **Identify the API surface.** Use `Config` / `@export_config` for portable recipes; use jsonargparse for known, typed CLI construction. Read `docs/explanation/configuration.md` and a matching unit test before changing behavior. - Done when: the change is located in one owner and preserves existing recipes. 2. **Implement with the trust boundary intact.** Only instantiate `class_path` from trusted local config. Keep the depth limit and `ConfigError` messages precise; artifact-path validation belongs to inference manifest resolution, not `physicalai.config`. Do not introduce a generic loader that executes untrusted YAML. - Done when: accepted and rejected inputs have explicit outcomes. 3. **Check downstream construction.** Round-trip an `@export_config` instance through `Config.from_instance(...).save()` and `instantiate()`; verify an existing Runtime YAML and Studio's use of `physicalai.config` still work. - Done when: round-trip and dependent construction tests pass without copying a config module into Studio. 4. **Update canonical documentation** under `docs/how-to/config/`, `docs/explanation/configuration.md`, or `docs/reference/config-schema.md` when the public contract changes. - Done when: an installed-package user can understand the new behavior without reading source. ## Verify ```bash uv run pytest tests/unit/config/ -q prek run ruff-check --all-files python3 .github/scripts/skills/agent_skills.py validate ```