Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →tsinghua-fib-lab/easypaper-interactive-metadata-build
Interactively build EasyPaper metadata from a research-materials folder using Claude Code's file-investigation tools. Output is JSON consumable by the `easypaper-paper-from-metadata` skill.
Passed 15 of 20 test cases with the skill, and 5 of 20 without it.
and 13 more in the full report
13 fixed4 passed either way4 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
I have a research directory at `/projects/graph_neural_nets` containing code, plots, and bib files. I would like to interactively build the metadata for my paper rather than auto-generating it silently. Please help me explore the folder, draft the main narrative fields (title, core hypothesis, method description, datasets used, and experiment details), identify image and table assets with relative paths, process my BibTeX citations, and validate the JSON object against the `PaperGenerationRequest` structure before saving it to disk.
| Checked | Without skill | With skill |
|---|---|---|
| Presents mode selection options cold, warm-start, and refine. | Missed | Passed |
| Waits for explicit user choice before proceeding to folder exploration. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill when the user has a folder of research materials and wants Claude to co-author the metadata in conversation instead of running the one-shot SDK pipeline silently.
The two paths are complementary:
| Path | Driver | Interaction | Use when |
|---|---|---|---|
SDK one-shot (generate_metadata_from_folder) | EasyPaper internal LLM | None | Batch, CI, or fast autonomous extraction |
| This skill | Claude Code | High | Single high-value paper, atypical folder, or ambiguity requiring user input |
Both paths produce metadata that can be consumed by easypaper-paper-from-metadata.
The final JSON must validate as PaperGenerationRequest and include:
title, idea_hypothesis, method, data, experimentsreferencesmaterials_rootfigurestablestemplate_path, style_guide, target_pages, generation flags, and output settingsEach figure/table object follows the SDK shape:
{
"id": "fig:h<12hex>",
"caption": "...",
"description": "...",
"section": "",
"file_path": "relative/posix/path/to/asset.png",
"wide": false,
"auto_generate": false,
"generation_prompt": null
}
Tables use the same shape with tab: ids.
test cases passed, vs 5 / 20 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.