comment-on-post

name	comment-on-post
description	Writes a LinkedIn comment on SOMEONE ELSE's post on behalf of an author. Loads the author's DIP, analyzes the post, picks a formula, and checks source grounding. Use when you need to comment on another person's post.
disable-model-invocation	true
argument-hint	["Author"]
allowed-tools	Read, Grep, Glob

Comment on Someone's Post

Write a comment on someone else's LinkedIn post on behalf of an author.

Comments are NOT posts (the core principle)

Adrian Vega's 500-post study + the LinkedIn 360Brew penalty: a comment that sounds like a mini-post is flagged by the algorithm as automation / AI spam. Symptoms of a bad comment:

• Post structure: hook → body → CTA
• Multiple paragraphs
• Em dashes, jagged rhythm, lists
• "Three insights:" / "Here's the thing:"
• Ends with a question (an AI marker and bad tone)

A good comment sounds like a chat reaction:

• 1–3 sentences, ONE thought
• A concrete reply to a concrete point in the original post
• Can be grammatically loose (fragments, no period at the end)
• Does NOT end with a question
• Optionally: a meme, an image, a short reaction
• Can start with a lowercase letter or a conjunction ("and yeah, ...", "but at the same time...")

Voice Vocabulary in comments

Filler words and tics from _Voice-Vocabulary/[Author]-Voice-Vocabulary.md ("Filler Words / Tics" and "Sentence Starters") are used in COMMENTS more than in posts — a comment is closer to spoken speech. This is what makes the comment recognizable as the author rather than a generic LinkedIn blob.

Do not use in comments:

• Voice-only markers explicitly flagged in the Voice Vocabulary as "not for posts" (e.g., "well there", "y'know" — spoken speech, not written)
• AI markers from detect-ai/SKILL.md Level 4F/4G/4H/4I (A-vs-B, false vulnerability, three lessons, generic CTA)

Reference: Knowledge-Base/02-LinkedIn-Algorithm/AI-Detection-Reference.md

Input

• $0 — author name (Seva, Kirill, Plurio)
• Conversation context — text/link to the post being commented on

Required files

1. Author's DIP: Knowledge-Base/01-Identity-Profiles/[Author]-DIP.md

   • Mapping: Seva → Seva-Ustinov-DIP.md, Kirill → Kirill-Kasimskiy-DIP.md, Plurio → Plurio-AI-DIP.md

2. Comments Engine: Knowledge-Base/06-Comments-Engine/Comments-Engine.md

Pipeline

Step 1: Load DIP

Extract from the DIP:

• Tone of Voice + Speech Patterns
• "They NEVER" — forbidden patterns
• Expertise Clusters — topics the author can speak to authoritatively
• Stories Bank — concrete cases and facts for grounding

Step 2: Post analysis

1. Read the original post
2. Determine:
   • What is the post about? Which expertise cluster does it match?
   • Is there room to add a unique perspective?
   • What tone is appropriate? (serious, light, provocative)
   • If the topic is OUTSIDE the author's clusters — warn the user

Step 3: Pick a formula

From the Comments Engine, Type 2 (comment on someone else's post):

Situation	Formula
Add your own experience	"We did something similar + [result]"
Clarify / extend	"This + [missing piece]"
Ask a smart question	"Curious about [specific aspect]"
Contrarian take	"Interesting — our data shows [different]"

Step 4: Write the comment

Rules:

• Minimum 15 words (for the 2.5× algorithmic weight)
• Optimal 30–80 words
• First person (I/my)
• Specifics from the DIP Stories Bank
• No hashtags
• No emoji (unless the author uses them in the DIP)
• Tone = DIP, but slightly softer (we're a guest here)

Step 5: Source Grounding

For EACH factual claim in the comment:

Category	Action
Verified fact — has a concrete source	Keep
Grounded inference — direct consequence of verified facts	Keep, don't embellish
Thin ground — author's topic but no concrete case	Rephrase: question or known fact
Bullshit — sounds plausible but invented	Remove or replace with a question

Red flag: >50% of claims are thin ground / bullshit → rewrite.

Step 6: Validation

• Sounds like the author? (DIP tone check)
• Adds value? (not an empty comment)
• 15+ words?
• No forbidden patterns from DIP "They NEVER"?
• First person?
• Not selling or pitching?
• Doesn't duplicate the original post's content?
• No "not X but Y" constructions?
• No "I've seen..." without specifics?
• No generic validation ("Great post!", "So true!", "This is the shift")?
• Source grounding passed?
• Doesn't sound like a mini-post (no hook→body→CTA, no multiple paragraphs)?
• Does NOT end with a question?
• Uses Voice Vocabulary elements (author's fillers/tics, where appropriate)?
• No AI markers detect-ai Level 4F–4I (A-vs-B opener, false vulnerability, three lessons, generic CTA)?

Output

Return:

1. Comment — ready-to-publish text
2. Source grounding table — each claim → source → category
3. Alternative variant — a different angle/formula (if appropriate)

Anti-Patterns (FORBIDDEN)

• "Great post!" / "Love this!" / "So true!"
• Quoting a line from the post + "this is the line/shift/key"
• Echoing metrics/phrases from the post back to the author
• "I've seen..." / "I've watched..." without specifics
• "not X but Y" constructions
• Transformational pivots ("it stops being X and becomes Y")
• Self-promotion on someone else's post
• "That's exactly how we designed it" and variants
• Selling/pitching in comments