| name | oob-perf-analysis |
| description | Generate and analyze T1/T2/R roofline reports for PyTorch OOB workloads comparing Intel XPU and NVIDIA CUDA. Use when working with eager profiling artifacts, per-model reports, fleet summaries, graph consistency, or XPU-vs-CUDA software efficiency analysis. |
OOB Perf Analysis Skill
Analyzes OOB eager-mode performance using T1/T2/R roofline methodology to compare XPU and CUDA software efficiency.
Core outputs (all under agent_space_xpu/, git-ignored):
- Per-model markdown reports under
agent_space_xpu/reports/<session>/models/
- Fleet summary under
agent_space_xpu/reports/<session>/summary_eager_inference.md
- Graph consistency report under
agent_space_xpu/reports/<session>/graph_consistency_eager_inference.md
- Insights summary under
agent_space_xpu/reports/<session>/insights_summary.md
References
Usage Modes
Per-Model Report Mode
User has raw artifacts and wants model-level T1/T2/R analysis.
Follow: methodology.md → inputs.md → per-model-report.md
Fleet Summary Mode
User wants fleet-wide comparison, model scorecards, op ranking, or projection-quality aggregation.
Follow: fleet-summary.md (references graph-consistency.md for Section 7)
Graph Consistency Mode
User wants to compare CUDA vs XPU computational graphs.
Follow: graph-consistency.md
Insights Mode
Per-model reports and fleet summary exist; user wants a concise developer-facing summary.
Follow: insights.md
Troubleshooting Mode
User asks why R is abnormal, why traces disagree, or why results look suspicious.
Follow: troubleshooting.md
Operating Rules
- Read
methodology.md before computing any metric.
- Read
inputs.md before accessing any artifact.
- Report structure is deterministic — same inputs produce same section layout.
If a calling workflow explicitly requires a skill marker, append this exact literal final line:
Custom skills applied: oob-perf-analysis.
