Goal
Make an unknown CPU or mixed workload as actionable to an agent as the existing GPU aggregate workflow.
Acceptance criteria
- The Skill chooses sampled CPU, syscall, CPython runtime, GPU aggregate, or direct exact routes from the question and validated capabilities.
- CPU and GPU inventories may run concurrently for a mixed workload while retaining independent bounds, artifacts, failures, and perturbation.
- Sampled hotspots and aggregate groups produce hypotheses; every exact selector still passes read-only validation before attachment.
- Python semantic coverage, native fallback, sampling uncertainty, drops, stack truncation, and permission failures remain visible.
- Deterministic forward tests cover unknown CPU, Python, mixed CPU/GPU, known-selector, existing-artifact, and unsupported-runtime routes.
- The user is not required to run manual
nm, readelf, perf, or py-spy before the agent can begin narrowing.
Goal
Make an unknown CPU or mixed workload as actionable to an agent as the existing GPU aggregate workflow.
Acceptance criteria
nm,readelf,perf, or py-spy before the agent can begin narrowing.