skilless.ai-writing（熟练的 ai 写作）

Writing Skill

Produce professional, research-backed written content.

Use when user needs to: write reports, articles, emails, documentation, briefs, memos, or any content that requires research and structured presentation.

When to Use

• User asks to "write", "draft", "compose", "create content"
• User needs a report, analysis, or documentation
• User wants an article, blog post, brief, or memo
• Any writing task that benefits from research-backed data and citations

Core Principle

Every piece of writing produced by this skill must be research-first, evidence-based, and professionally structured. Do not write from general knowledge alone — always gather real data, verify claims, and cite sources.

Writing Process

Step 1 — Understand the Brief

Before writing anything, clarify:

• What type of content? Report, article, email, documentation, memo, etc.
• Who is the audience? Executive, technical, general public, etc.
• What tone? Formal, conversational, technical, persuasive, etc.
• What is the desired length? Brief (1-2 pages), standard (3-5 pages), comprehensive (5+ pages)
• Are there specific points or questions to address?

If any of the above is unclear, use the question tool to batch-ask the user before proceeding.

Step 2 — Research (Invoke Research Skill)

Always research before writing. Invoke skilless.ai-research to gather data:

1. Determine research depth based on the writing task:

- Quick email or short summary → L1 (1-2 searches) - Standard article or comparison → L2 (3-5 searches, 2-4 page reads) - Full report or deep analysis → L3 (5+ searches, 5+ page reads, cross-source fact-check)

1. Let the research skill handle tool selection — it owns all CLI tools (search, web reader, video transcript extractor, media converter) and will choose the right ones based on the task.
2. Fact-check all key claims following the research skill's fact-check protocol:

- Cross-verify key data from 2-3 independent sources - Note contradictions and resolve or flag them - Mark unverified claims explicitly

1. Collect citation data for every fact, statistic, and claim:

- Source title, URL, publication date - Specific data points with original phrasing - Author/organization credibility

Step 3 — Outline

Before writing prose, create a structured outline:

• Start with the conclusion / key finding
• Organize supporting sections logically
• Identify where each data point and citation will go
• Ensure every major claim has a source assigned

Step 4 — Write

Follow the report structure defined below. Write clearly, concisely, and professionally.

Step 5 — Review

Before delivering:

• Verify every citation is real and correctly referenced
• Check that the executive summary accurately reflects the full content
• Ensure no unsupported claims remain
• Confirm the output follows the format rules

Report Structure

All reports and substantial writing follow this structure. Adjust section depth based on content length.

1. Executive Summary

Always lead with this. A self-contained summary that a busy reader can use without reading the full report.

Must include:

• Key finding or recommendation — the single most important takeaway, in 1-2 sentences
• Critical data points — the 3-5 most important numbers or facts, with citations
• Context — why this matters, in 1-2 sentences
• Recommended action (if applicable) — what the reader should do

Length: 150-300 words for standard reports, 50-100 words for short briefs.

2. Background / Context

• Why this research was conducted
• Scope and limitations
• Key definitions or assumptions

3. Findings / Analysis

The detailed body of the report. Organize by theme, question, or comparison dimension.

Requirements:

• Every claim backed by specific data — not vague statements like "many users prefer X", but "67% of surveyed users preferred X [3]"
• Concrete numbers — prices, percentages, dates, version numbers, performance metrics
• Source citations on every data point — inline [1], [2], etc.
• Comparisons use tables when data fits single-line cells
• Contradictions addressed explicitly — "Source A reports X [1], while Source B reports Y [2]. The discrepancy likely stems from [reason]. Source A is considered more reliable because [reason]."

4. Conclusion / Recommendations

• Summarize findings (not a copy of the executive summary — this is more detailed)
• Provide actionable recommendations with rationale
• Note limitations and areas needing further investigation

5. Sources

Full citation list at the end:

## Sources

[1] [Title](URL) — Key data point used
[2] [Title](URL) — Key data point used
[3] [Title](URL) — Key data point used

Writing Standards

Data Quality

• Use specific numbers, not vague language. Instead of "significant growth", write "42% year-over-year growth [2]"
• Include dates and versions. Instead of "the latest version supports X", write "v3.2 (released March 2025) supports X [4]"
• Attribute opinions. Instead of "X is considered the best", write "Gartner ranked X as the market leader in their 2025 report [5]"
• Flag uncertainty. If data is limited, write "Based on available data (single source), X appears to be Y — further verification recommended"

Tone Matching

Adapt tone to the user's request and audience:

If the user does not specify tone, default to professional/formal for reports and analysis.

Length Calibration

• Do not pad with filler — every sentence should carry information
• Do not over-compress to the point of losing important nuance
• If the user asks for a "brief" or "summary", aim for 300-500 words
• If the user asks for a "full report" or "detailed analysis", aim for 1000-3000 words
• When in doubt, use the question tool to ask about desired depth

Handling Different Content Types

Asking the User

Use your environment's built-in question/ask tool when needed. Batch related questions into a single call.

Output Format

Strict portable Markdown only. Must render correctly in any Markdown editor (GitHub, Obsidian, Typora, VS Code, etc.).

Rules

1. No HTML tags — no <br>, <div>, <table>, <sub>, <sup>, or any HTML whatsoever
2. Table cells must be single-line plain text — no line breaks, no nested lists, no multi-line content
3. If content does not fit single-line table cells, use a list instead of a table
4. Blank lines before and after headings, tables, code blocks, and block quotes
5. Fenced code blocks only (triple backticks) — no indented code blocks
6. No trailing spaces for line breaks — use separate paragraphs instead

Formatting Toolkit

Source Citations

Inline: [1], [2], etc. — placed immediately after the data point they support.

At the end of every document:

## Sources

[1] [Title](URL) — Key data point used
[2] [Title](URL) — Key data point used

Cross-References

• Research & data gathering → Invoke skilless.ai-research for all search, web reading, video transcript extraction, and media operations. The research skill owns all CLI tools (search, web, youtube, ffmpeg) and defines research depth levels (L1–L3)
• Brief unclear? → Invoke skilless.ai-brainstorming to clarify scope, audience, and direction before starting the writing process