Blog Writer

Blog Writer

Write, refine, and adapt blog posts with a tech-first focus (not tech-only). Supports personal blogs, internal email blogs, WeChat official account articles, and newsletters.

Core Workflow

Phase 1: Context Resolution (MANDATORY before writing)

Resolve all parameters before writing. The goal is minimum friction, maximum signal — infer everything possible from the user's message and environment, then ask only what is genuinely ambiguous.

1a. Auto-Infer from User Message and Environment

Extract as much as possible without asking:

| Parameter | How to infer | |---|---| | Language | Match the user's message language. Chinese message → Chinese blog. English message → English blog. If ambiguous (e.g., mixed-language message with no clear dominant language), ask. | | Topic | Extract from the user's message. "写一篇关于 Redis 缓存的博客" → topic is Redis caching. | | Format / channel | Default to personal technical blog unless the user says "内部邮件", "newsletter", "公众号", or similar. | | Audience | Default to fellow engineers unless the user specifies otherwise. | | Tone / style | Infer from topic type (see style mapping below). Can be overridden later. | | Source materials | If the user referenced files, directories, URLs, or @-mentions, those are the source materials. Read them in Phase 2. | | Output path | See "File Output Rules" below. |

Style mapping (default tone based on topic type):

| Topic type | Default style | |---|---| | Sharing a project / tool / open-source library | Pragmatic, pain-point-first | | Technical principle / architecture deep-dive | Rigorous, evidence-based | | Framework / tool comparison | Analytical, judgment-driven | | Incident postmortem / lessons learned | Honest, retrospective | | Personal reflection / non-technical | Casual, show the journey | | Concept explainer | Analogy-driven, light teaching tone |

1b. Identify Gaps and Confirm

After extracting what you can, assess what's missing. There are three possible outcomes:

Case 1 — Everything is clear: Topic, source materials (or clearly not needed), and output path are all known. Proceed directly to Phase 2. Show a one-line summary of what you're about to write before starting research:

> 收到。中文个人技术博客，关于 ，务实简洁风格，面向工程师。开始研读素材。

Case 2 — Only topic is missing: The user said "写篇博客" without specifying what about. Ask for the topic with numbered choices:

> 博客主题是什么？ > > 1. 分享一个项目 / 工具 / 开源库 > 2. 解析某个技术原理或架构设计 > 3. 对比多个方案 / 框架 / 工具 > 4. 总结踩坑经验 / 故障复盘 > 5. 个人思考 / 非技术话题 > 6. 其他 (请描述)

After the user answers, proceed to Phase 2 (do NOT ask follow-up questions about style/audience/format — use defaults).

Case 3 — Multiple things are ambiguous: Present a confirmation block that lists all inferred values and highlights the gaps. The user can confirm, override, or fill in missing values in a single reply.

Example (Chinese):

> 我理解的参数如下，请确认或修改： > > - 语言: 中文 > - 主题: （请补充） > - 渠道: 个人技术博客 > - 读者: 工程师同行 > - 风格: 务实简洁 > - 素材: 无（写出来会比较泛，建议提供代码/文档/链接） > - 存放位置: ./blogs/ > > 回复确认，或修改任意项。

Example (English):

> Here's what I've inferred — confirm or adjust: > > - Language: English > - Topic: (please specify) > - Channel: Personal tech blog > - Audience: Fellow engineers > - Style: Pragmatic and concise > - Materials: None (output may be generic — consider sharing code/docs/links) > - Output path: ./blogs/ > > Reply to confirm, or change any item.

Rules for the confirmation block:

• Show ALL parameters (inferred + missing) in one place — the user gets a complete picture.
• Mark missing items with (请补充) / (please specify).
• The user can reply with just "ok" / "确认" to accept all defaults, or override specific items like "风格改成轻松有趣" / "style: casual".
• After confirmation, proceed to Phase 2. Do NOT ask further questions.

1c. Handling "Just Write It"

If the user says "just write" / "直接写" without providing a topic or enough context:

• If topic is missing: you MUST ask — a blog without a topic cannot be written.
• If topic is known but no source materials: proceed with a warning that output will be generic. Do NOT block.

Phase 2: Research & Preparation

1. Read AGENTS.md / project conventions: If the output directory (or its parent) contains an AGENTS.md, CLAUDE.md, or similar convention file, read it. Adopt its style rules, formatting conventions, and structural preferences for the blog. Project-level rules override this skill's defaults where they conflict.
2. Read source materials: If the user pointed to code, README files, project directories, or URLs — read them thoroughly. Understand the technical substance before writing.
3. Study reference articles: If the user provided example articles for style reference, analyze their structure, tone, paragraph rhythm, and formatting conventions.
4. Identify the narrative arc: Plan a "why should I care" hook → problem exploration → solution → honest assessment arc before writing.
5. Select angles: If source materials are extensive (large README, many files), identify the 2–3 most interesting angles and propose them to the user for selection before writing.

Phase 3: Writing

Structure

• Open with a hook: A specific problem, a relatable frustration, a counterintuitive claim, or a concrete scenario. NEVER open with "随着 XX 技术的发展……" or "在当今数字化时代……"
• Use informative headings: H2/H3 headings should tell the reader what they'll learn. Avoid generic headings like "背景介绍" or "总结".
• End with substance: Summarize the core conclusion (don't just repeat the body). Include actionable next steps, links, or an honest assessment of limitations.

Format-Specific Rules

Personal blog (default):

• 2000–5000 words (can be shorter if user requests)
• Full narrative arc from pain point to solution
• Include code examples where relevant (must be runnable, not pseudocode)
• For non-technical topics, replace code with concrete examples, data, or scenarios
• Discuss limitations honestly — this builds trust
• After the draft, provide: 3 alternative titles + 1 SEO meta description

Internal email blog / newsletter:

• 1000–2000 words, 5-minute read target
• Use a concise title with a clear core message
• Numbered sections (一、二、三) for easy scanning
• Faster pace: shorter paragraphs, less setup, more conclusions
• End with clear action items (links, install commands)
• Focus on ONE core value proposition — don't try to cover everything

WeChat official account:

• 1500–3000 words
• Mobile-friendly: shorter paragraphs, more whitespace
• Can use bold for emphasis on key sentences
• Consider adding a TL;DR at the top

Style Rules

• Write in the resolved language and stay consistent throughout.
• Keep code, proper nouns, and product names in their original form.
• If Chinese:
• Add spaces around English technical terms: 使用 Redis 缓存 not 使用Redis缓存
• Use Arabic numerals: 3 个节点 not 三个节点
• Use "我" not "笔者"
• If English:
• Prefer short, clear sentences and concrete claims over abstract phrasing
• Avoid passive-heavy paragraphs unless needed for clarity
• For other languages, ask whether the user wants formal or conversational tone.
• Have opinions. Make judgments. Avoid empty hedging.
• Vary sentence length.
• One idea per paragraph, max 5 lines per paragraph.

Anti-Patterns to Avoid

• README regurgitation: Don't copy-paste feature lists, installation steps, or API docs. Summarize by use case, link to the README for details.
• Exhaustive enumeration: Don't list every feature. Group by scenario, highlight core value for each group.
• Empty emphasis: Don't bold text that isn't actually important.
• Filler phrases: No "无需赘述", "相信读者已经了解", "如有不足，欢迎指正".
• Buzzword soup: No "赋能", "落地", "闭环", "抓手".

Phase 4: Output & Review

File Output Rules

Format: Markdown (.md) by default. Only use a different format if the user explicitly requests it.

File naming: Lowercase English, hyphen-separated, descriptive. Examples:

• redis-caching-strategies.md
• .md
• skill-management-internal-email.md (suffix for alternate versions of the same topic)

Output path resolution (in priority order):

1. User-specified path: If the user explicitly said where to put the file (e.g., "放到 my-blogs/docs/ 下"), use that path.
2. AGENTS.md / project conventions: If the target directory or workspace has an AGENTS.md that specifies where blog files go (e.g., "文章保存为 Markdown 文件，放在 blogs/ 目录下"), follow that convention.
3. Existing blog directory: If the workspace has a blogs/ directory with existing .md files, use it.
4. Current directory: If none of the above apply, save to the current working directory.

Create parent directories if they don't exist (mkdir -p).

Post-Draft Deliverables

After the draft:

1. Explain choices briefly: Why this opening, this tone, this structure — 2–3 sentences, not a paragraph.
2. Provide alternatives (for personal blog format): 3 alternative titles + 1 SEO meta description.
3. Invite iteration: Ask if the user wants to adjust tone, length, depth, or emphasis.

Editing an Existing Article

When the user asks to edit/refine an existing article (not write from scratch):

• Preserve the author's voice — don't rewrite from scratch.
• Point out logical gaps, unclear explanations, or missing context.
• Suggest specific cuts and rewrites with reasoning.
• Check for "README-ification" and suggest rewrites if found.

Handling Edge Cases

• User gives a topic but no source material: Proceed but warn that output may be generic. For deeply technical topics, suggest what materials would help (e.g., "如果能提供代码仓库或 README，技术细节会更准确").
• User wants multiple formats from the same content: Write the primary format first, then adapt. Don't write both simultaneously — the structure and pacing differ.
• User requests bilingual output: Draft in one primary language first, then create a localized adaptation.
• User provides a very long README or codebase: Don't try to cover everything. Identify the 2–3 most interesting angles and propose them to the user for selection.
• Conflicting style rules: AGENTS.md in the target directory takes precedence over this skill's defaults. If AGENTS.md says "use 我们" but the user says "use 我", follow the user's explicit instruction.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jie-meng
• Source: jie-meng/mythril-agent-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.