Skip to content
AI & Agents
Command

/easypaper-metadata-build

Interactively build EasyPaper metadata from a research-materials folder using Claude Code's file-investigation tools. Output is JSON consumable by `/easypaper-paper-from-metadata`.

BOOST
From plugin
agentsociety
1.4k5 skills5 commands
Install
$ npx -y skills add tsinghua-fib-lab/agentsociety --agent claude-code

How it fires

How this command gets triggered: by you, by Claude, or both.

  • Fires itselfClaude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/easypaper-metadata-build

Context preview

What this command does when you run it.

Interactively build EasyPaper metadata from a research-materials folder using Claude Code's file-investigation tools. Output is JSON consumable by `/easypaper-paper-from-metadata`.

Command definition

easypaper-metadata-build.md

Interactively build EasyPaper metadata from a research-materials folder using Claude Code's file-investigation tools. Output is JSON consumable by `/easypaper-paper-from-metadata`.

Use this command when the user wants Claude to co-author metadata instead of running the SDK one-shot `ep.generate_metadata_from_folder(...)` pipeline silently.

Execution Contract

1. Present the three modes and wait for an explicit choice:

  • `[1] cold` — Claude walks the folder alone, no Python SDK call.
  • `[2] warm-start` (recommended) — call `ep.generate_metadata_from_folder(materials_root, max_figures=12, max_tables=12, vision_enrich_figures=False)` once, then review/edit every field with the user.
  • `[3] refine` — load an existing metadata JSON, validate it with the inline checks below, then walk the user through fixes and warnings.

2. Ask for the corresponding input:

  • `cold` / `warm-start`: `materials_root`; resolve it and verify it is an existing directory.
  • `refine`: existing metadata JSON path; verify it exists.

3. Ask for optional style/template settings:

  • `style_guide`: target venue or style guide.
  • `template_path`: path to the LaTeX template zip if known. If relative, interpret it relative to the saved metadata file location during generation.

4. Hand off to the `easypaper-interactive-metadata-build` skill for the detailed 6-phase workflow. 5. Save to `<materials_root>/easypaper_metadata.json` by default and tell the user to run `/easypaper-paper-from-metadata` next.

Inline Validation Rules

Before saving, validate the in-memory metadata with public SDK/Pydantic plus explicit checks:

  • `PaperGenerationRequest.model_validate(metadata_dict)` succeeds.
  • `title`, `idea_hypothesis`, `method`, `data`, and `experiments` are non-empty strings.
  • `materials_root` is set to a resolved directory for folder-derived metadata.
  • `figures[].file_path` and `tables[].file_path` are relative POSIX paths under `materials_root`.
  • Figure ids use the `fig:` prefix; table ids use the `tab:` prefix.
  • Figure and table counts are within the selected caps.
  • Empty `references` is a warning, not a hard failure.

Output Rules

  • `materials_root` is an absolute path to the source folder.
  • Figure/table paths are relative POSIX paths under `materials_root`.
  • Do not convert folder-derived figure/table paths to absolute paths in the saved metadata.
  • `template_path` may be omitted or written relative to the metadata file location when it is part of the same portable project.

$ARGUMENTS

Read more
Ships withagentsociety

AgentSociety 2 is a modern, LLM-native agent simulation platform designed for social science research and experimental design. It provides a flexible framework for creating and managing intelligent agents in simulated environments.

Get the whole plugin

Other commands on agentsociety.