菜单
Parse one candidate Darshan-style trace and preserve evidence for later comparison.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
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_candidate_trace |
| title | Parse One Candidate Trace |
| description | Parse one candidate Darshan-style trace and preserve evidence for later comparison. |
Use this skill when the expert is responsible for the candidate side of an HPC I/O regression comparison.
Always parse the candidate 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 decide root cause in this expert. Return normalized candidate metrics to the parent so comparison and attribution happen after both branches return.
