论文写作
- 作者仓库星标 190
- 作者更新于 实时读取
- 作者仓库 skilless.ai
- 领域
- 写作
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @BrikerMan · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 需简单配置
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 底层运行要求
- 无特殊要求
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- Shell 执行
- 网络行为
- 仅限本地
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: skilless.ai-writing
description: Help produce written content by researching topics and gathering references. Use when user needs…
category: 写作
runtime: 无特殊运行时
---
# skilless.ai-writing 输出预览
## PART A: 任务判断
- 适用问题:文章、文案、发言稿、润色或结构化表达。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“When to Use / Core Principle / Writing Process”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于文章、文案、发言稿、润色或结构化表达,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“When to Use / Core Principle / Writing Process”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、执行终端命令、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件、执行终端命令;主要在本地完成;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文没有稳定的斜杠命令要求。安装验证后通常全局生效,直接在对话里点名这个 Skill 并描述任务即可。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件、执行终端命令。
先用一个小任务确认它会围绕“When to Use / Core Principle / Writing Process”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: skilless.ai-writing
description: Help produce written content by researching topics and gathering references. Use when user needs…
category: 写作
source: BrikerMan/skilless.ai
---
# skilless.ai-writing
## 什么时候使用
- skilless.ai-writing 是 BrikerMan 在 BrikerMan/skilless 适合处理文章、文案、润色、翻译、总结和结构化表达,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答…
- 面向文章、文案、发言稿、润色或结构化表达,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「When to Use / Core Principle / Writing Process」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件、执行终端命令;主要在本地完成;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "skilless.ai-writing" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> When to Use / Core Principle / Writing Process
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> 无特殊运行时 | 读取文件、写入/修改文件、执行终端命令 | 主要在本地完成
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} 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:
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)
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.
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
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:
| Audience | Tone | Example phrasing |
|---|---|---|
| Executive / decision-maker | Formal, concise, action-oriented | "We recommend adopting X based on 35% cost reduction [1]" |
| Technical team | Precise, detailed, specification-focused | "Latency reduced from 120ms to 45ms (p99) under 10K concurrent connections [3]" |
| General public | Clear, accessible, jargon-free | "This means your battery will last about twice as long as the previous model" |
| Internal memo | Direct, brief, bullet-point heavy | "Action needed: approve budget by Friday" |
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
| Content type | Research depth | Structure | Key focus |
|---|---|---|---|
| Full report / analysis | L3 | Full report structure (exec summary through sources) | Data depth, cross-verification, actionable recommendations |
| Article / blog post | L2-L3 | Hook, body sections, conclusion, sources | Engaging narrative, supported claims, clear takeaways |
| Documentation / guide | L2 | Overview, steps/sections, examples, references | Accuracy, completeness, practical examples |
| Email / memo | L1-L2 | Key point first, supporting details, action items | Brevity, clarity, specific asks |
| Comparison / evaluation | L2-L3 | Exec summary, comparison table, detailed analysis, recommendation | Fair representation, concrete metrics, clear winner/tradeoffs |
Asking the User
Use your environment's built-in question/ask tool when needed. Batch related questions into a single call.
| Situation | Example question(s) |
|---|---|
| Content type or audience unclear | "To write this effectively: 1) Who is the target audience? 2) What tone — formal report or casual article? 3) Desired length?" |
| Scope too broad to cover well | "This topic is quite broad. Would you like me to: 1) Cover all aspects at a high level 2) Focus deeply on [specific area] 3) Something else?" |
| Research found contradictory data | "My research found conflicting data on [topic]: Source A says X, Source B says Y. How should I handle this in the report? 1) Present both with analysis 2) Go with the more credible source 3) Flag for further investigation" |
| Missing critical information | "I could not find reliable data on [specific point]. Should I: 1) Proceed without it and note the gap 2) Try alternative search approaches 3) Adjust the scope?" |
| Draft complete — follow-up | After delivering the content: "What would you like me to do next? 1) Expand a specific section 2) Adjust tone or format 3) Research additional aspects 4) Create a shorter summary version 5) Something else" |
Output Format
Strict portable Markdown only. Must render correctly in any Markdown editor (GitHub, Obsidian, Typora, VS Code, etc.).
Rules
- No HTML tags — no
<br>,<div>,<table>,<sub>,<sup>, or any HTML whatsoever - Table cells must be single-line plain text — no line breaks, no nested lists, no multi-line content
- If content does not fit single-line table cells, use a list instead of a table
- Blank lines before and after headings, tables, code blocks, and block quotes
- Fenced code blocks only (triple backticks) — no indented code blocks
- No trailing spaces for line breaks — use separate paragraphs instead
Formatting Toolkit
| Element | Usage |
|---|---|
| Bold | Key conclusions, important numbers, recommendations |
Lists (- or 1.) |
Pros/cons, steps, supporting points |
✅ ❌ ⚠️ |
Supported / not supported / caution |
> quote block |
Direct quotes from sources |
`inline code` |
Technical terms, tool names, commands |
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-researchfor 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-brainstormingto clarify scope, audience, and direction before starting the writing process
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核