メニュー
Parse one baseline Darshan-style trace and preserve evidence for later comparison.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Compare scientific variables with units, ranges, and quality caveats.
Inspect dataset structure before analysis or visualization.
Evaluate data quality from observed structure and summary evidence.
Choose visual checks that match the inspected data.
Route dataset questions to structure, analysis, or visualization experts.
Choose a plot that confirms numeric conversion integrity.
| name | parse_one_baseline_trace |
| title | Parse One Baseline Trace |
| description | Parse one baseline Darshan-style trace and preserve evidence for later comparison. |
Use this skill when the expert is responsible for the baseline side of an HPC I/O regression comparison.
Always parse the baseline trace with the assigned tool before making claims. Return a compact evidence block containing the original path, runtime, rank count, read/write byte totals, read/write timing, metadata timing, collective and independent write counts, transfer-size hints, and any parser warnings.
Do not compare against the candidate trace in this expert. Return the normalized baseline metrics to the parent so the parent can merge both sides.
