academic-paper-style-rules

name	academic-paper-style-rules
title	Academic Paper Writing with Strict Style Rules
description	Write academic papers (.txt, .md) with enforced style constraints: banned words, voice rules, reference verification, IEEE structure.
version	1.0.0
author	Ion Agent
license	MIT
metadata	{"ion":{"tags":["Research","Paper Writing","Style Rules","Academic Writing","Plain Text","Citations","arXiv"],"category":"research","related_skills":["research-paper-writing","arxiv","humanizer"],"requires_toolsets":["terminal","files","web_extract"]}}

Academic Paper Writing with Strict Style Rules

Compose academic research papers in plain text formats (.txt, .md) while adhering to rigorous style constraints. This skill spans the entire workflow: outlining the structure, drafting with style compliance, validating references, and performing pre-delivery checks.

Activate this skill when: The user requests a paper with explicit style constraints (prohibited vocabulary, voice requirements, formatting limitations) and expects plain-text output.

Prefer research-paper-writing instead when: Producing LaTeX papers targeting ML conferences (NeurIPS, ICML, ICLR) with automated citation management.

---

Quick Start

1. Parse style rules from the user's request and build a checklist
2. Outline the paper structure (IEEE-format sections)
3. Draft the paper while respecting all style constraints
4. Batch-verify every reference (arXiv IDs, author lists)
5. Run compliance scans (search for banned words across ALL text, including headings)
6. Deliver the paper alongside a verification summary

---

Style Rule Enforcement

Extracting Rules

When the user supplies style rules, produce an explicit checklist:

STYLE CHECKLIST:
- Voice: [we/I/one/they]
- Banned words: [word1, word2, ...]
- Banned punctuation: [em dashes, semicolons, etc.]
- Required structure: [IEEE/custom/listed sections]
- Citation format: [arXiv IDs/DOIs/numbered refs]
- Definitions: [formal with propositions/informal]
- Claim boundaries: [precise with qualifiers/approximate]

Common Banned Words (AI Tells)

These words are hallmarks of AI-generated text. Inspect ALL text, including section headings:

Banned	Replace With
delve	explore, examine, investigate
landscape	ecosystem, field, area, domain
tapestry	remove entirely
crucial	essential, necessary, required, critical
furthermore	also, additionally, (or just start next point)
moreover	also, and, (or restructure)
in conclusion	To summarize, We conclude, (or just conclude)
it is important to note	remove entirely, state the fact directly
in the realm of	in, within, for
aforementioned	this, these, the (or name it directly)
notably	specifically, in particular, (or remove)
particularly	specifically, especially, (or remove)

Checking ALL Text Locations

Critical pitfall: Banned words often hide in section headings and titles, not just body text.

# Check entire file including headings
grep -in 'word1\|word2\|word3' paper.txt

# Check for em dashes
grep -c '—' paper.txt  # Should be 0

Example: "II. BACKGROUND / A. The Current Benchmark Landscape" contains "landscape". Fix: "A. The Current Benchmark Ecosystem".

Voice Enforcement

When "we" voice is required:

• Use "we" for all actions: "we propose", "we argue", "we define"
• Avoid passive: "it is shown" → "we show"
• Avoid "the author(s)": use "we"
• Avoid "this paper": use "we"

---

Reference Verification

The Problem

Plain-text papers lack programmatic citation tooling (no .bib files). References are composed from memory, which leads to:

• Incorrect arXiv IDs (paper title correct, ID wrong)
• Incomplete or inaccurate author lists
• Wrong publication years
• Mismatched venues

Verification Workflow

1. Draft the paper with references from memory
2. Collect all arXiv IDs cited in the paper
3. Batch-verify via web_extract (max 5 URLs per call):
   • Title matches the arXiv ID
   • Author list matches
   • Year matches submission history
4. Correct discrepancies before delivery

Batch Verification Pattern

# Group references into batches of 5
batch_1 = [
    "https://arxiv.org/abs/2311.12983",  # GAIA
    "https://arxiv.org/abs/2310.06770",  # SWE-bench
    "https://arxiv.org/abs/2308.03688",  # AgentBench
    "https://arxiv.org/abs/2307.13854",  # WebArena
    "https://arxiv.org/abs/2403.07718",  # WorkArena
]
# web_extract each batch, compare results against paper

Common Errors Found

Error Type	Example	Detection
Wrong arXiv ID	2405.06001 cited as LoRA paper (actually LLMC)	Extract page, check title
Wrong first author	"Gasse et al." vs actual "Drouin et al."	Check author list on page
Missing arXiv ID	Conference ref without arXiv	Search for arXiv version
Wrong year	Submission year vs publication year	Check submission history

When arXiv Page Doesn't Show Authors

Some arXiv pages don't render author lists in extracted text. Fallback:

1. Search for "paper title" arxiv authors
2. Check Semantic Scholar API
3. Check the HTML version: https://arxiv.org/html/{id}v{version}

---

Paper Structure

IEEE-Style Template (Standard)

Title

Author Name(s)
Affiliation
Correspondence: email

ABSTRACT
[Single paragraph, 150-250 words, states problem/approach/results]

INDEX TERMS
[Comma-separated, 5-8 terms]

I. INTRODUCTION
[Problem, motivation, contributions as bullet list, paper organization]

II. BACKGROUND / RELATED WORK
[Existing work, benchmarks, prior art, positioning]

III-V. MAIN SECTIONS
[Framework, metrics, protocol, evaluation, experiments]

VI. RELATED WORK (if separate from background)
[Detailed comparison with prior work]

VII. LIMITATIONS
[Honest assessment: computational cost, scope, assumptions]

VIII. CONCLUSION
[Summary of contributions, future work, invitation to community]

REFERENCES
[Numbered list, each with arXiv ID or DOI]

Word Count Targets

Confirm with user before writing:

• Short: 3,000-4,000 words
• Standard: 4,000-6,000 words
• Long: 6,000-8,000 words

---

Pre-Delivery Checklist

Run ALL checks before delivering the paper:

# 1. Word count
wc -w paper.txt

# 2. Em dashes
grep -c '—' paper.txt  # Must be 0

# 3. All banned words (including headings)
grep -in 'delve\|landscape\|tapestry\|crucial\|...' paper.txt

# 4. Reference count
grep -c '^\[' paper.txt  # Should match expected count

# 5. Section headers present
grep -E '^(I\.|II\.|III\.|...)' paper.txt

# 6. Verify "we" voice (if required)
grep -c ' we ' paper.txt  # Should be high

Verification Summary Template

After completing the paper, provide:

VERIFICATION RESULTS:
- Word count: X (within Y-Z range)
- Em dashes: 0
- AI tells: None found
- References: N (all verified)
- Sections: All required sections present
- Voice: "we" used throughout
- File: /path/to/file.txt

---

Pitfalls

1. Section headings contain banned words: Always grep the entire file, not just body text
2. arXiv ID mismatch: Paper title can be correct while ID is wrong; verify both
3. Author list drift: Well-known papers can have wrong author lists in memory; always verify
4. Year confusion: Submission year vs publication year vs camera-ready year differ
5. Em dash variants: Check for both — (Unicode) and -- (ASCII double hyphen)
6. Transitional phrases: "Furthermore" and "Moreover" often creep in during long writing sessions; check at end
7. Voice consistency: Switch between "we" and passive voice mid-paragraph; do a focused check

---

Related Skills

• research-paper-writing: LaTeX papers for ML conferences with programmatic citations
• arxiv: Search arXiv for papers to cite
• humanizer: Strip AI-isms from text

Reference Files

• references/verification-case-studies.md: Real examples of citation errors caught during paper writing, with detection and fix patterns (5 error types documented with specific arXiv ID corrections)