Abstract

Abstract — Scholarly and technical communication is frequently weakened by defensive writing: an authorial posture characterized by pre-emptive apologies, repetitive disclaimers, stacked modal hedges, and negative self-deprecation. While intended to ward off hypothetical criticism, defensive habits dilute the empirical strength of claims and obscure core scientific contributions.

Anti-Defensive Writing introduces an agentic skill specification that systematically diagnoses and eliminates defensive cognitive overhead while strictly preserving methodological validity, empirical scope, and ethical/legal boundaries. By replacing defensive disclaimers with positive scope definitions and calibrating uncertainty to observed evidence, this framework enables researchers and professionals to communicate findings with authoritative clarity, structural focus, and maximum signal-to-noise ratio.

The complete operational rulebook and multilingual instructions are codified in SKILL.md (comprising English Core Rules, English Extended Guidelines, Chinese Core Rules, and Chinese Extended Guidelines).

Core Principles

The framework establishes four fundamental pillars for rigorous, claim-forward academic communication:

Technical Plans and SOPs

In Chinese technical plans, SOPs, reports, and presentations:

Remove redundancy and admonitory tone while preserving technical meaning, applicability conditions, prohibitions, and operational purpose. Do not change the operation to make the text sound more concise or specialized.

• State the plan directly: Give the procedure, parameters, prerequisites, and findings. Remove commentary about the writing only when it adds no substantive information. Keep phrases such as “用于阐明机制” when they state a real research purpose.
• Name technical prerequisites: Replace vague wording such as “在条件允许时” with concrete prerequisites supported by the draft, for example, “标定前提：……”. Keep necessary conditions expressed by “仅在……时” or “若欲……需要……”, or rewrite them equivalently. Do not replace them with an unconditional action; identify missing conditions when they are unspecified.
• Use operational language: State the required handling directly and remove admonitory wording such as “切记”, while retaining necessary prohibitions and technical conclusions. Keep “不可互相替代” when it expresses a real restriction. Use “两者独立记录并平行分析” only when the procedure calls for it; this does not convey non-interchangeability.
• Choose precise verbs: Simplify redundant phrasing such as “进行……的测量” to “测量……”. “测量” and “表征” are valid technical verbs. Use “标定”, “解耦”, “淬灭”, or “钳定” only when they describe the actual operation; do not substitute a different operation to sound more specialized.

The Chinese guide includes the Chinese rules and examples.

Comparative Analysis

The table below contrasts conventional defensive drafts against revisions generated under the Anti-Defensive Writing paradigm across standard scholarly contexts.

Rewrite Pipeline

Revising text with Anti-Defensive Writing follows a formal 5-stage transformation pipeline:

┌────────────────────────────────────────────────────────┐
│               Input: Defensive Academic Draft          │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│  Stage 1: Functional Diagnosis                         │
│  Classify defensive units: Unnecessary Disclaimer vs.  │
│  Necessary Scope vs. Methodological Constraint         │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│  Stage 2: Disclaimer Pruning                           │
│  Strip uninformative disclaimers and preemptive        │
│  apologies that contribute no empirical value          │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│  Stage 3: Positive Projection                          │
│  Convert negative denials ("Not X") into               │
│  explicit positive definitions ("Focuses on Y")        │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│  Stage 4: Uncertainty Calibration                      │
│  Collapse stacked modal hedges ("may potentially");    │
│  ground uncertainty in concrete observational limits   │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│  Stage 5: Structural Realignment                       │
│  Anchor paragraphs with decisive topic sentences;      │
│  isolate true constraints to dedicated sections        │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│          Output: Assertive, High-Signal Paper          │
└────────────────────────────────────────────────────────┘

Quickstart

Deployment Footprint

The installation is engineered to be minimal, hermetic, and zero-overhead. Only a single specification file is deployed into your target skills environment:

<skills-directory>/
└── anti-defensive-writing/
    └── SKILL.md

Zero Runtime Dependencies: The installer streams SKILL.md directly. It does not clone repository archives, nor does it install README files, scripts, images, test suites, or auxiliary configurations. It requires no Git, Python, Node.js, compilation toolchains, or package managers.

Supported Environments

Note: ~ denotes your user home directory. Relative paths resolve from the execution directory. For project-level installations, execute commands directly from the project root.

Installation via Shell (macOS / Linux / WSL)

Choose one command corresponding to your target agent:

Standard Codex (~/.agents/skills)

curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.sh | sh -s -- --dest "$HOME/.agents/skills"

Legacy Codex (~/.codex/skills)

curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.sh | sh -s -- --dest "$HOME/.codex/skills"

Claude Code (~/.claude/skills)

curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.sh | sh -s -- --dest "$HOME/.claude/skills"

For project-level installation, pass a relative destination such as --dest ".agents/skills", --dest ".codex/skills", or --dest ".claude/skills".

Installation via PowerShell 7 (Windows)

Open PowerShell 7 (pwsh). Configure $skillsDir to your target directory, then execute:

$skillsDir = "$HOME/.agents/skills"
$installSkill = [scriptblock]::Create((Invoke-RestMethod 'https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.ps1'))
& $installSkill -Dest $skillsDir

For alternative environments, set $skillsDir to "$HOME/.codex/skills" or "$HOME/.claude/skills". For project-level scope, specify ".agents/skills" from your project root.

Upgrades & Version Pinning

Existing installations remain untouched unless explicitly updated using --force (Shell) or -Force (PowerShell). Atomic replacement guarantees that a failed network download will never corrupt an existing installation.

# Shell: Force update to latest main
curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.sh | sh -s -- --dest "$HOME/.agents/skills" --force

# Shell: Pin to specific commit, tag, or branch
curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/install.sh | sh -s -- --dest "$HOME/.agents/skills" --ref <commit-hash>

# PowerShell: Force update
& $installSkill -Dest $skillsDir -Force

# PowerShell: Pin reference
& $installSkill -Dest $skillsDir -Ref <commit-hash> -Force

Manual Installation

To install manually without automated scripts, create anti-defensive-writing inside your chosen skills directory and save the raw SKILL.md directly:

mkdir -p ~/.agents/skills/anti-defensive-writing
curl -fsSL https://raw.githubusercontent.com/Kiterlin/anti-defensive-writing/main/SKILL.md -o ~/.agents/skills/anti-defensive-writing/SKILL.md

Agent Interface

Once installed, invoke the skill directly inside your agent session:

Codex CLI

$anti-defensive-writing Revise these paragraphs. Remove unnecessary defensive language and preserve the evidence and necessary qualifications.

Claude Code CLI

/anti-defensive-writing Revise these paragraphs. Remove unnecessary defensive language and preserve the evidence and necessary qualifications.

Interactive Web Interfaces (ChatGPT / Claude / Gemini)

For web-based LLM sessions without local file system access, attach or paste SKILL.md directly as custom system instructions or context prompt.

Recommended Prompt Patterns

$anti-defensive-writing Refactor the following academic introduction. Eliminate apologetic preambles, replace negative exclusions with positive scope definitions, and ensure the topic sentence leads directly with our core finding.

$anti-defensive-writing Review this contributions section. Remove self-undermining modesty, preserve all true experimental scope limits, and present our empirical trade-offs without defensive justification.

$anti-defensive-writing Rewrite this rebuttal paragraph. Directly answer the reviewer's concern using observed experimental evidence. Do not concede imaginary weaknesses or use apologetic phrasing.

Case Studies

Empirical before-and-after transformations across common academic genres are cataloged in examples/:

• Academic Paper Introduction: Transforming a hesitating, apologetic opening in platform governance into an evidence-forward hook.
• Methods & Contributions: Articulating experimental contributions and computational bottlenecks with calibrated objectivity.
• Grant Proposal (NSF/NIH Style): Reframing tentative project aims into high-conviction, feasibility-grounded scientific objectives.

Methodological Guardrails

[!IMPORTANT] Anti-Defensive Writing is NOT ungrounded hype. It enforces strict scientific integrity constraints:

1. No Hallucination of Evidence: The skill strictly forbids fabricating facts, metrics, baselines, or mechanisms to make a sentence sound stronger.
2. No Causal Overreach: Observational associations must never be transformed into causal assertions.
3. Preservation of Essential Boundaries: Limitations critical to validity, ethics, safety, legal compliance, or replication must be preserved and clearly stated in their appropriate sections.

Repository Layout

anti-defensive-writing/
├── SKILL.md                         # Authoritative skill specification (Rules of truth)
├── skill/anti-defensive-writing/
│   └── SKILL.md                     # Hermetic single-file installable payload
├── README.md                        # English documentation & benchmark overview
├── README.zh-CN.md                  # Simplified Chinese documentation
├── install.sh                       # POSIX installer (macOS / Linux / WSL)
├── install.ps1                      # Windows PowerShell 7 installer
├── skill.json                       # Agent package manifest & metadata
├── examples/                        # Academic case studies & qualitative evaluations
│   ├── academic-introduction.md     # Paper introduction case study
│   ├── methods-and-contributions.md # Methodology & contributions case study
│   └── grant-proposal.md            # Research proposal case study
├── assets/                          # Architecture schematics & figures
│   ├── anti-defensive-writing-en.png
│   └── anti-defensive-writing-zh-CN.png
├── tests/                           # Installer verification & CI regression tests
└── LICENSE                          # MIT License

The mirror located at skill/anti-defensive-writing/SKILL.md is kept strictly identical to the root SKILL.md for tools requiring a directory-based skill format. When using third-party installers, point to skill/anti-defensive-writing rather than the repository root.

Citation

If you utilize Anti-Defensive Writing in your academic workflow, manuscripts, or agentic systems, please cite this work:

@software{antidefensivewriting2026,
  author       = {Sunlight and Contributors},
  title        = {Anti-Defensive Writing: Calibrated Assertiveness & Positive Scope for Academic and Technical Prose},
  year         = {2026},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/Kiterlin/anti-defensive-writing}}
}

License

This project is licensed under the MIT License.