一键在 Manus 中运行任何 Skill

scientific-data-parser

星标1

分支0

更新时间2026年3月17日 00:01

Parse specialized scientific data formats (e.g., HDF5, NetCDF, FITS, mzML, CIF, PDB) or unfamiliar raw data files. Use when the user wants to process complex scientific files into AI-Ready formats.

安装

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。

在 Manus 中运行

来源

oNya685

oNya685/NanoData

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Scientific Data Parsing SOP

Scientific data often contains a mix of structured numerical arrays and unstructured metadata/remarks. Follow this workflow to ensure ZERO data loss.

Phase 1: Exploration & Dependency Management

Never use read_file on raw data.
Text vs Binary:
- For text-based formats (CIF, PDB, XML, JSON, CSV, unfamiliar txt): Use the read_file_lines tool (e.g., lines 1-100) to understand the schema and locate metadata blocks.
- For binary formats (HDF5 .h5, NetCDF .nc, FITS .fits), do NOT read them as text.
Install Dependencies: If the format requires a specific library, use python_exec to install it first.
- HDF5 -> !pip install h5py
- NetCDF -> !pip install netCDF4
- FITS/Astronomy -> !pip install astropy
- Crystallography (CIF) -> !pip install pymatgen

Phase 2: Parser Script Generation with Strict Feedback Loop

When writing the Python parsing script (via python_exec), you must adhere to the Zero Data Loss Rule:

Read skills/scientific-data-parser/scripts/template_science.py for the standard architecture.
Your script must proactively catch any unparsed items, unmatched regex groups, or trailing metadata/remarks.
Raise Warnings: If the script finds extra columns, floating text, or undocumented metadata, it MUST print a warning to stderr or stdout in this format: [UNPARSED_WARNING] File: <path>, Line/Key: <location>, Content: <content>

Phase 3: Evaluate & Iterate

After running the script, carefully read the output from python_exec.
If you see [UNPARSED_WARNING], it means your script missed valuable scientific context.
Self-Correction: Rewrite the script to explicitly extract those missed items, map them to appropriate JSON fields (like metadata or operator_notes), and run it again until no warnings are generated.
Output the final data using the standard dual-output format (.md and .jsonl).

同仓库更多 Skills

同仓库

clawhub

oNya685/NanoData

Search and install agent skills from ClawHub, the public skill registry.

2026-03-171

cron

oNya685/NanoData

Schedule reminders and recurring tasks.

2026-03-171

data-cleaning

oNya685/NanoData

End-to-end data preprocessing pipeline to transform raw files (CSV, Excel, Parquet) into AI-Ready assets. Use this skill when the user requests to clean data, handle missing values, fix data types, remove duplicates, standardize formats, or prepare datasets for model training. Includes automatic profiling and dataset card generation.

2026-03-171

data-to-text

oNya685/NanoData

Convert structured data (CSV/Parquet/Excel) into semantic, readable text documents or experimental reports for LLM fine-tuning. Use when the user wants to transform tabular data into natural language narratives, generate training corpus, or create AI-Ready text documents from datasets. Output is Markdown format ready for further processing.

2026-03-171

github

oNya685/NanoData

Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.

2026-03-171

multimodal-augmentation

oNya685/NanoData

Generate images for existing SFT datasets (JSON/JSONL) to build multimodal fine-tuning data; use when you need to add image prompts, descriptions, and saved image files for each QA pair or conversation, via SubAgents + image_generate tool.

2026-03-171

name	scientific-data-parser
description	Parse specialized scientific data formats (e.g., HDF5, NetCDF, FITS, mzML, CIF, PDB) or unfamiliar raw data files. Use when the user wants to process complex scientific files into AI-Ready formats.

Scientific Data Parsing SOP

Scientific data often contains a mix of structured numerical arrays and unstructured metadata/remarks. Follow this workflow to ensure ZERO data loss.

Phase 1: Exploration & Dependency Management

Never use read_file on raw data.
Text vs Binary:
- For text-based formats (CIF, PDB, XML, JSON, CSV, unfamiliar txt): Use the read_file_lines tool (e.g., lines 1-100) to understand the schema and locate metadata blocks.
- For binary formats (HDF5 .h5, NetCDF .nc, FITS .fits), do NOT read them as text.
Install Dependencies: If the format requires a specific library, use python_exec to install it first.
- HDF5 -> !pip install h5py
- NetCDF -> !pip install netCDF4
- FITS/Astronomy -> !pip install astropy
- Crystallography (CIF) -> !pip install pymatgen

Phase 2: Parser Script Generation with Strict Feedback Loop

When writing the Python parsing script (via python_exec), you must adhere to the Zero Data Loss Rule:

Read skills/scientific-data-parser/scripts/template_science.py for the standard architecture.
Your script must proactively catch any unparsed items, unmatched regex groups, or trailing metadata/remarks.
Raise Warnings: If the script finds extra columns, floating text, or undocumented metadata, it MUST print a warning to stderr or stdout in this format: [UNPARSED_WARNING] File: <path>, Line/Key: <location>, Content: <content>

Phase 3: Evaluate & Iterate

After running the script, carefully read the output from python_exec.
If you see [UNPARSED_WARNING], it means your script missed valuable scientific context.
Self-Correction: Rewrite the script to explicitly extract those missed items, map them to appropriate JSON fields (like metadata or operator_notes), and run it again until no warnings are generated.
Output the final data using the standard dual-output format (.md and .jsonl).