Comment Draft

Comment Composition and Tone Control

Purpose

Generate appropriate comment drafts based on article content, comment section context, user-specified tone, and identity. Output options for user approval before posting.

Scope

What This Skill Does

• Analyze article/post content for comment relevance
• Consider existing comment sentiment and discussions
• Generate 1-3 comment drafts matching specified tone
• Respect word limits and platform conventions
• Flag potential risks (controversial topics, sensitive phrasing)
• Provide rationale for each draft option
• Support multiple languages based on content language

What This Skill Does NOT Do

• Execute any app operations (no tapping, scrolling)
• Post comments (use comment-posting-workflow)
• Read articles or comments (expects input from other skills)
• Make final decisions (user selects which draft to use)

Inputs

Required

Context Object

context:
  article:
    title: string
    summary: string  # 2-3 sentences
    key_points: list  # Main points to potentially reference
    language: string  # Content language

  comment_section:
    sentiment_summary: string  # From comment-reading skill
    top_comments: list  # Notable existing comments
    common_questions: list  # Unanswered questions
    controversies: list  # Debate points

  user_goal: string  # What user wants to achieve
  # Examples:
  # - "Share my perspective on AI safety"
  # - "Ask a clarifying question"
  # - "Support the author's point"
  # - "Offer a counterpoint respectfully"

Tone Options

Optional Parameters

Outputs

Comment Draft Output

comment_drafts:
  generated_at: timestamp
  input_summary:
    article_topic: string
    requested_tone: string
    identity: string

  drafts:
    - draft_id: 1
      content: string
      tone_match: float  # 0-1, how well it matches requested tone
      length: int  # Character count
      language: string

      strategy: string  # How this comment approaches the topic
      # Examples:
      # - "Adds personal experience to support main point"
      # - "Asks clarifying question about methodology"
      # - "Offers alternative perspective with evidence"

      references:  # What from context it builds on
        - type: string  # "article_point" / "existing_comment" / "user_goal"
          source: string

      risks:
        level: "none" | "low" | "medium" | "high"
        concerns: list  # Specific concerns if any
        # Examples:
        # - "Might be seen as dismissive of original post"
        # - "References controversial figure"
        # - "Could start heated debate"

    - draft_id: 2
      content: string
      # ... same structure

    - draft_id: 3
      content: string
      # ... same structure

  recommendation:
    suggested_draft: int  # Which draft is recommended
    reason: string

Primary Workflow

Step 1: Analyze Context

1. Understand article topic and stance
2. Note key points that could be referenced
3. Review comment section sentiment
4. Identify opportunities:
   - Unanswered questions to address
   - Points needing support
   - Gaps in discussion
   - Misconceptions to clarify

Step 2: Map Tone to Style

Based on requested tone, determine:

Professional:
  - Vocabulary: technical terms, precise language
  - Structure: clear thesis, supporting evidence
  - Avoid: slang, emoji, excessive enthusiasm

Casual:
  - Vocabulary: everyday words, contractions
  - Structure: conversational flow
  - Allow: light emoji, personal pronouns, anecdotes

Supportive:
  - Opening: appreciation ("Great point about...")
  - Body: build on existing ideas
  - Closing: encouragement

Constructive_critical:
  - Opening: acknowledge merit ("I see the logic here")
  - Body: respectful disagreement with reason
  - Closing: open to discussion

Curious:
  - Frame as questions, not statements
  - "I'm wondering...", "Could you elaborate..."
  - Show genuine interest

Step 3: Incorporate Identity

Identity affects:
  - How commenter introduces themselves (if at all)
  - What knowledge/experience they can reference
  - Credibility signals

Examples:
  Identity: "AI researcher"
  → Can reference research experience
  → Should use accurate terminology
  → Should not overclaim expertise

  Identity: "curious reader"
  → Ask questions, not make assertions
  → Can express learning journey
  → Should not pretend expertise

  Identity: "industry practitioner"
  → Can share real-world examples
  → Practical perspective
  → Avoid academic jargon

Step 4: Generate Draft Variations

For drafts_count drafts:

Draft 1: Direct response to main article point
  - Most straightforward approach
  - Clear connection to content

Draft 2: Engage with existing comments
  - Build on or respond to notable comments
  - Join the conversation

Draft 3: Unique angle
  - Add new perspective
  - Ask unexpected question
  - Share relevant experience

Step 5: Apply Constraints

For each draft:
  1. Check length against max_length
     - If over, condense without losing meaning

  2. Check forbidden_topics
     - Remove any forbidden references
     - Flag if removal changes meaning significantly

  3. Check must_include
     - Ensure all required points are addressed

  4. Language check
     - Ensure output language matches specification
     - For Chinese content, use appropriate formality

Step 6: Risk Assessment

Evaluate each draft for:

HIGH RISK:
  - Directly contradicts/attacks author
  - References sensitive political topics
  - Makes unverifiable claims
  - Could be seen as spam/promotional

MEDIUM RISK:
  - Strong opinion on controversial topic
  - Disagrees with popular comments
  - References ongoing debates

LOW RISK:
  - Expresses opinion with hedging
  - Asks questions
  - Shares personal experience only

NO RISK:
  - Supportive comments
  - Neutral observations
  - Simple questions

Step 7: Recommend Best Draft

Selection criteria:
  1. Best tone match
  2. Lowest risk (unless user wants to be provocative)
  3. Most relevant to user's goal
  4. Most likely to get engagement

Heuristics

Length Guidelines by Platform

Opening Patterns by Tone

Culture-Aware Phrasing

Chinese platforms (WeChat, Weibo):

• More indirect criticism preferred
• "个人观点" (personal opinion) as hedging
• Emoji and stickers more common
• Reference to "学习了" (learned something) for appreciation

English platforms:

• Direct but polite
• "I think" / "In my experience" for hedging
• Emoji usage varies by platform

Failure Modes & Recovery

1. Insufficient Context

Symptom: Not enough information to write relevant comment.

Recovery:

• Request more context from user
• Generate very general comment with disclaimer
• Suggest what additional info would help

2. Conflicting Requirements

Symptom: Tone and identity don't match (e.g., "casual" + "formal expert").

Recovery:

• Prioritize tone (explicit request)
• Adjust identity expression
• Note conflict in output

3. Impossible Constraints

Symptom: Max length too short for meaningful comment.

Recovery:

• Note limitation
• Provide shortest possible version
• Suggest longer version if constraints relaxed

4. Highly Controversial Topic

Symptom: No way to comment without taking controversial stance.

Recovery:

• Flag as high risk
• Offer "question only" alternative
• Suggest user reconsider engaging

5. Language Mismatch

Symptom: User wants to comment in different language than content.

Recovery:

• Can generate, but flag potential reception issues
• Suggest matching content language

6. All Drafts Too Similar

Symptom: Different approaches converge on same comment.

Recovery:

• Be more creative with angles
• Explicitly try contrasting strategies
• Note if topic limits variation

Tooling (MCP)

This skill does NOT use MCP tools.

It is a pure text generation skill that:

• Receives context from other skills
• Outputs draft text for user approval
• Feeds selected draft to comment-posting-workflow

Examples

Example 1: Professional Supportive Comment

Input:

context:
  article:
    title: "Why AI Agents Will Transform Software Development"
    summary: "Author argues AI agents can autonomously handle routine coding tasks, freeing developers for creative work."
    key_points:
      - "Agents can write boilerplate code"
      - "Human oversight still essential"
      - "5-10 years to mainstream adoption"

  comment_section:
    sentiment_summary: "Generally positive, some skepticism about timeline"
    top_comments:
      - "Finally someone gets it - agents are tools, not replacements"

  user_goal: "Support the author's balanced view"

tone: professional
identity: "Software engineer with AI experience"
max_length: 200

Output:

comment_drafts:
  drafts:
    - draft_id: 1
      content: "Excellent analysis. As someone who's integrated AI agents into our development workflow, I can confirm the productivity gains are real—but so is the need for human review. The 5-10 year timeline feels right given current capabilities."
      tone_match: 0.95
      length: 198
      strategy: "Validate from personal experience"
      risks:
        level: "none"

    - draft_id: 2
      content: "This captures the nuance often missing from AI discussions. The agent-as-tool framing is crucial. In my team's experience, the biggest wins come from reducing context-switching, not replacing engineers entirely."
      tone_match: 0.90
      length: 196
      strategy: "Add complementary insight"
      risks:
        level: "none"

    - draft_id: 3
      content: "Building on the oversight point—we've found that agents actually improve code review quality because reviewers can focus on architecture and logic instead of style and boilerplate. Win-win."
      tone_match: 0.85
      length: 188
      strategy: "Extend specific argument"
      risks:
        level: "low"
        concerns: ["Might seem like humble-bragging about team practices"]

  recommendation:
    suggested_draft: 1
    reason: "Most direct support for author, balanced, mentions real experience"

Example 2: Constructive Criticism (Chinese)

Input:

context:
  article:
    title: "大模型将在两年内取代程序员"
    summary: "作者认为大模型发展迅速，两年内将完全取代初级程序员工作"
    language: "zh-CN"

  user_goal: "Respectfully disagree with timeline"

tone: constructive_critical
identity: "行业从业者"
language: "zh-CN"
max_length: 200

Output:

comment_drafts:
  drafts:
    - draft_id: 1
      content: "作者对技术发展的观察很敏锐，不过个人认为两年的时间线可能过于乐观。从实际落地来看，AI目前更像是提效工具而非替代方案。当然技术发展常有惊喜，拭目以待。"
      tone_match: 0.92
      length: 89
      strategy: "承认观点后委婉质疑"
      risks:
        level: "low"

    - draft_id: 2
      content: "思路很有启发性。想请教：这个两年预测是基于哪些技术节点的判断？从我接触的项目看，复杂业务逻辑的处理仍是难点，很好奇作者怎么看这块。"
      tone_match: 0.88
      length: 94
      strategy: "以提问方式引导反思"
      risks:
        level: "none"

    - draft_id: 3
      content: "有启发。不过历史上类似预测（如'5年内实现无人驾驶'）往往低估了工程落地的复杂度。AI辅助编程是趋势，'取代'可能需要更长时间观察。个人看法，仅供参考。"
      tone_match: 0.85
      length: 98
      strategy: "引用历史类比"
      risks:
        level: "medium"
        concerns: ["可能被认为在否定作者观点"]

  recommendation:
    suggested_draft: 2
    reason: "以提问形式最委婉，避免直接对抗"

Example 3: Casual Curious Comment

Input:

context:
  article:
    title: "My Experience Using AI Agents for a Month"
    summary: "Personal blog post about trying various AI agents for daily tasks"

  comment_section:
    common_questions:
      - "Which agent did you find best for coding?"

  user_goal: "Ask about the author's coding workflow"

tone: casual
identity: "curious reader"
max_length: 150

Output:

comment_drafts:
  drafts:
    - draft_id: 1
      content: "Super interesting write-up! I'm curious - for the coding tasks you mentioned, did you find any agent particularly good at debugging vs. writing new code? Trying to figure out which to try first 🤔"
      tone_match: 0.95
      length: 148
      strategy: "Specific follow-up question"
      risks:
        level: "none"

    - draft_id: 2
      content: "This is exactly what I needed to read! Quick q - were you using these agents with any specific IDE or just web interfaces? Trying to set up something similar 👀"
      tone_match: 0.90
      length: 142
      strategy: "Practical setup question"
      risks:
        level: "none"

    - draft_id: 3
      content: "Love the honest take! Do you think a complete beginner could follow a similar workflow, or is some coding background needed to make good use of these tools?"
      tone_match: 0.88
      length: 145
      strategy: "Accessibility question"
      risks:
        level: "none"

  recommendation:
    suggested_draft: 1
    reason: "Most directly addresses unanswered question in comments"

Key Reminders

1. Pure text generation - No app operations, just compose text
2. Multiple options - Always provide choices, let user decide
3. Match content language - Unless user explicitly wants different
4. Risk transparency - Always flag potential issues
5. Respect constraints - Length limits are hard limits
6. Tone consistency - Every word should match requested tone
7. Identity authenticity - Don't claim expertise the identity doesn't have
8. User goal alignment - Drafts should serve user's stated purpose