v0.15.1

Highlights

• Completes the Learning plane control surface for runtime-generated learning work.
• Adds runtime evaluation of compiled learning trigger rules and queues targeted learning requests.
• Allows targeted learning requests to start directly at the requested learning stage, such as Curator.
• Surfaces learning queue depth and learning status in millrace status show.
• Preserves skill revision evidence per stage request and labels learning stage requests as learning_request.

Verification

• uv run --with ruff ruff check src/millrace_ai tests
• uv run --with mypy mypy src/millrace_ai
• uv run pytest -q
• uv build
• uv run --with twine twine check dist/*
• Fresh Learning plane E2E: workspaces/learning-plane-e2e-003/e2e_report.json with overall_success: true