essay-writing

name

essay-writing

description

Plan academic manuscripts from a structured `Manuscript-preparation` folder without writing the manuscript body. Use when the task is to analyze `Draft`, `Guide-for-authors`, `References`, `Dur-team's-research-output`, optional `Sample-for-authors`, and `Supplementary-information.md`, create `Manuscript-preparation/Draft-preparation/Essay.md`, build a three-level reference system for later `initial-draft-writing`, generate `Draft-preparation/important-references.rdf`, generate `Draft-preparation/RDF-extraction-summary.md` when RDF sources exist, summarize `Supplementary-information.md` into `Draft-preparation/Sinformation-summary .md`, and reliably read multiformat evidence including office files, spreadsheets, diagrams, scientific project files, and images without turning folder labels such as `Dur-team's-research-output` into author-centered headings.

Essay Writing

Overview

Use this skill to understand the manuscript task, organize the evidence, and plan the writing strategy before any manuscript body is written. Create the planning artifacts inside Manuscript-preparation/Draft-preparation and stop there.

Treat folder names as routing metadata, not wording to copy into output. Use the materials in Dur-team's-research-output to foreground the user's prior results and evidence, but do not write headings or section labels such as authors' team, our team, or similar author-centered phrasing unless the user explicitly requests it.

Runtime Notes

Keep the core workflow self-contained in this file so runtimes that ignore agents/ can still use the skill.

Treat helper skills or format-specific tooling as optional accelerators, not hard dependencies. If doc, pdf, slides, or spreadsheet skills exist, use them when they improve reliability. Otherwise use the bundled scripts and the best available local tools.

When a runtime supports a skill base-directory variable, use {baseDir} to reference bundled scripts. In runtimes that do not expand {baseDir} automatically, resolve the script path relative to the folder that contains this SKILL.md file instead of assuming the workspace root.

Bundled Resources

• scripts/extract_rdf.py: Recursively find .rdf files, parse RDF/XML with Zotero and EndNote-friendly heuristics, resolve creator lists, containers, collections, notes, and attachment links, fall back to raw text when necessary, and emit a Markdown or JSON summary that the agent can reliably read.
• scripts/extract_multiformat_sources.py: Recursively find supported source files, extract best-effort readable text from office documents, spreadsheets, diagrams, scientific project files, PDFs, plain text files, and images, and emit a Markdown or JSON summary that the agent can reliably read.

Use the bundled scripts in this order when the corresponding files exist:

1. Run extract_multiformat_sources.py on References and Dur-team's-research-output to build the first-pass readable surface for all supported formats.
2. Run extract_rdf.py on any .rdf sources for structured bibliographic interpretation.
3. If the multiformat summary reports image:metadata-only, open the critical image files with a vision-capable tool before finalizing the plan.
4. If the multiformat summary reports a best-effort scientific-project parse for .opju or .prism, verify critical labels, values, or figure intent against exported images, tables, or attachments when available.

Folder Contract

Expect the user to place all source material under one folder named Manuscript-preparation.

Find these items case-insensitively by basename and allow the listed formats to vary:

• Dur-team's-research-output/
• References/
• Draft in .doc, .docx, .pdf, .md, .txt, .ppt, or .pptx
• Guide-for-authors in .doc, .docx, .pdf, .md, or .txt
• Supplementary-information.md
• Optional Sample-for-authors in .doc, .docx, .pdf, .md, or .txt
• Output folder Draft-preparation/ to be created if missing

Treat Dur-team's-research-output as a multiformat evidence folder. Its files may include, but are not limited to, .doc, .docx, .pdf, .md, .txt, .ppt, .pptx, .vsdx, .xls, .xlsx, .opju, .prism, .rdf, .png, .jpg, .jpeg, .tif, .tiff, and related nested-source files.

Search Dur-team's-research-output and References recursively. Do not stop at the first level. Treat every supported document found anywhere below those folders as in scope, including deeply nested files.

Treat the contents as follows:

• Treat References as the literature pool to be filtered into a three-level reference system.
• Treat Dur-team's-research-output as the user's prior results, preferred terminology, chronology, strongest directly reusable evidence, and source of reusable tables, figures, and supporting materials.
• Treat Draft as the source of the fixed manuscript title and the current abstract, keywords, or structural hints.
• Treat Guide-for-authors as the authoritative source for structure, formatting, length, section order, citation style, and submission constraints.
• Treat Supplementary-information.md as extra user instructions or context. Allow it to be empty.
• Treat Sample-for-authors as a style and structure reference only. Do not copy its wording.

If any required item is missing, stop and ask the user for the missing file or folder before planning.

Outputs

Create or update these artifacts under Manuscript-preparation/Draft-preparation:

• Essay.md
• important-references.rdf
• RDF-extraction-summary.md when any .rdf files were found
• Sinformation-summary .md when Supplementary-information.md is not empty

Do not write manuscript body text into Draft.

Non-Negotiable Rules

• Do not write or revise the manuscript body in Draft.
• Preserve the manuscript title exactly as written in Draft. Give title optimization suggestions only in a separate note.
• Ask for the target language if it is not clear from the user request or the existing manuscript context.
• Use the materials in Dur-team's-research-output to foreground the user's results, but do not turn the folder name into author-centered headings or wording.
• Build a three-level reference system: core references, chapter-specific references, and reserve references.
• Prefer supplied materials over web research. Use web research only to fill necessary gaps, verify unstable facts, or satisfy an explicit user request.
• Mark every web-sourced addition in red inside the working planning artifact and record every consulted URL in Manuscript-preparation/Information Sources.md.
• Do not invent bibliographic metadata, findings, figure meaning, or claims.

Format Handling

Use the bundled scripts to ensure the skill can actually read multiformat content before screening or outlining.

• Run python "{baseDir}/scripts/extract_multiformat_sources.py" <References> <Our-team-output> --output <temporary-summary> before ranking evidence.
• Run python "{baseDir}/scripts/extract_rdf.py" <References> <Our-team-output> --output Draft-preparation/RDF-extraction-summary.md when .rdf files exist.
• Treat .docx, .pptx, .xlsx, and .vsdx as directly readable package formats via the multiformat extractor.
• Treat .doc, .ppt, and .xls as legacy office formats; attempt automatic conversion first and fall back to best-effort text extraction if conversion is unavailable.
• Treat .opju and .prism as scientific project files; use the multiformat extractor first, then verify critical interpretation against exported figures, tables, or images when available.
• Treat .png, .jpg, .jpeg, .tif, and .tiff as first-class evidence. Use OCR or a vision-capable tool to inspect text-bearing images, charts, screenshots, and scanned pages.
• Treat .rdf as a readable source format, but always use the dedicated RDF extractor for structured interpretation instead of raw XML inspection.

If a file still remains only partially readable after the scripted pass, explicitly note the gap in the planning package instead of pretending it was fully understood.

Three-Level Reference System

Organize the literature into these layers:

1. Core references Use a reasonable count for the manuscript type. Prioritize papers that directly define the topic, provide the core theory or method base, establish the main comparison baseline, or deliver representative conclusions that initial-draft-writing should lean on heavily.
2. Chapter-specific references Assign targeted references to major sections such as introduction, related work, methods, results framing, discussion, and limitations. Use these to strengthen section-level relevance instead of relying only on one global paper list.
3. Reserve references Keep moderately relevant literature that may help with counterarguments, updated background, supplementary evidence, or expanded discussion. Keep these available but do not plan to use them deeply by default.

For each selected reference, capture why it was selected, what section or sections it supports, and what evidence role it plays.

important-references.rdf

Write important-references.rdf as a lightweight RDF/XML manifest of the selected core references.

For each selected core reference, include only metadata that is actually supported by the source materials, such as:

• title
• identifier or DOI when known
• source path or source URI
• short selection reason
• planned section usage
• evidence role such as definition, method basis, comparison baseline, discussion anchor, or key comparative study

Do not invent missing fields. Omit unknown metadata instead.

Workflow

1. Inspect the package

Locate the Manuscript-preparation folder and create Draft-preparation if it does not already exist.

Walk References and Dur-team's-research-output recursively before deciding which source files are most relevant. Do not assume the useful files live directly under those two folders.

Run the multiformat extractor across References and Dur-team's-research-output before outlining so that office files, diagrams, spreadsheets, scientific project files, PDFs, and images are not silently skipped.

If any .rdf files are present, generate Manuscript-preparation/Draft-preparation/RDF-extraction-summary.md with the bundled RDF script before outlining. Read that generated summary as part of the source review. If it lists linked local attachments that exist on disk, read those linked PDFs, DOCX files, images, or other source documents immediately after reviewing the RDF summary.

Extract these anchors before outlining:

• exact manuscript title from Draft
• current abstract and keywords from Draft
• target article type, section order, length, and citation style from Guide-for-authors
• any extra constraints or terminology from Supplementary-information.md
• any domain chronology, method lineage, preferred naming, figure logic, or data cues from Dur-team's-research-output

2. Summarize supplementary information

If Supplementary-information.md contains meaningful content, create Manuscript-preparation/Draft-preparation/Sinformation-summary .md and write a concise, cleaned summary of the actionable instructions, terminology notes, chronology, exclusions, and user preferences.

Do not copy the supplementary note verbatim unless a quotation is necessary.

3. Build the literature and evidence map

Perform layered screening over References and the evidence-bearing items in Dur-team's-research-output.

Use a full shallow scan first. Read titles, abstracts, keywords, the multiformat summary, the RDF summary, visible metadata, image content, and obvious relevance cues across the whole evidence set before deciding what is core.

Then assign each useful source to one of the three levels and capture:

• why the source matters
• what section it supports
• whether it is central or optional
• what evidence type it provides
• whether it supplies reusable images, tables, diagrams, or data for the later initial draft

4. Write important-references.rdf

Create Manuscript-preparation/Draft-preparation/important-references.rdf from the selected core references.

Use it as a compact handoff package for later initial-draft-writing, not as a dump of all references.

5. Write Essay.md

Create Manuscript-preparation/Draft-preparation/Essay.md and write the outline there.

Make Essay.md detailed enough to directly drive initial-draft-writing. Include at least:

• fixed manuscript title
• optional title suggestions in a separate note
• target language and style assumptions
• article type and author-guide constraints
• section-by-section structure
• planned evidence map for each major section
• core references with reasons
• chapter-specific references grouped by section
• reserve references with short notes on when to revisit them
• writing strategy notes that help the later initial draft
• notes on where images, tables, charts, or diagrams may be cited, adapted, or newly generated later

Before finishing Essay.md, make sure the planning language does not use headings such as authors' team case studies. Refer to the user's prior results neutrally, for example through topic-based or method-based phrasing.

Integrity Checks

Before delivering the planning package, verify all of the following:

• Draft body text was not modified
• Draft-preparation/Essay.md exists and is coherent
• important-references.rdf exists and contains only selected core references
• RDF-extraction-summary.md exists when .rdf files were present
• Sinformation-summary .md exists when supplementary information was non-empty
• multiformat evidence was actually screened instead of only reading plain-text sources
• the title from Draft is unchanged inside the planning package
• the folder name Dur-team's-research-output was used only as routing metadata, not as author-centered output wording
• no fabricated bibliographic metadata or unsupported claims were introduced

Example Requests

• Use $essay-writing to analyze my Manuscript-preparation folder, prepare Draft-preparation/Essay.md, and do not write the Draft document yet.
• Use $essay-writing to build a three-level literature system, generate important-references.rdf, and summarize Supplementary-information.md.
• Use $essay-writing to foreground the user's prior results without writing headings such as authors' team.