LinkedIn Marketing Skills

The linkedin-marketing bundle provides 11 specialized skills for LinkedIn content operations, designed for Claude Code and Codex. The skills follow a consistent draft → approval → publish workflow.

Available Skills

• linkedin-post-writer: Write viral posts.
• linkedin-comment-drafter: Draft comments on specific post URLs.
• linkedin-reply-handler: Reply to existing comments.
• linkedin-humanizer: Audit drafts, remove AI tells, score emoji density, and run AI detectors (includes --mode audit).
• linkedin-hook-extractor: Extract hook formulas from viral posts.
• linkedin-content-planner: Plan a week of content.
• linkedin-thread-monitor: Track which comments receive author replies.
• linkedin-engager-analytics: Analyze audience segmentation (likes/comments).
• linkedin-profile-optimizer: Audit or rewrite LinkedIn profiles.
• linkedin-employee-advocacy: Run advocacy programs across teams.
• linkedin-repurposer: Adapt content from other platforms (tweets, videos, blogs) into native LinkedIn posts.

The linkedin-humanizer uses a tiered approach to distinguish between actual AI leakage, bad writing style, and legitimate human literary patterns. This prevents the tool from 'flattening' good writing while still catching AI-generated content.

Tier 1: FORENSIC (Always On)

Targets pure leakage—patterns that no human writer produces. These are undeniable signals of AI origin.

• Examples: Internal tool tokens (e.g., oaicite, contentReference), model disclaimers (e.g., "As of my last update..."), unpopulated placeholders (e.g., [Your Name]), and extreme em dash frequency (3+ in <300 words).

Tier 2: STRICT (Default On)

Targets bad-style overlap—corporate-speak and filler that AI uses heavily. Removing these improves post quality even if the writer is human.

• Examples: Corporate synonyms (e.g., leverage, utilize, harness), filler adverbs (e.g., fundamentally, essentially), clichés (e.g., "in today's fast-paced world"), and negative parallelism ("X isn't Y, it's Z").

Tier 3: AESTHETIC (Opt-in Only)

Targets good-writing overlap—patterns AI uses that are also common in high-quality human writing. Using this mode blindly can flag legitimate authors as AI.

• Examples: Single em dashes, the 'rule of three' (tricolons), passive voice, specific adjectives (e.g., robust, vibrant), and curly quotes.
• When to use: Only for audiences that are extremely paranoid about AI detection (e.g., academic forums, Wikipedia editors).

The LinkedIn Humanizer is designed to be destructive to remove AI tells, but it must avoid destroying the user's unique 'voice.' Certain patterns are classified as voice signatures rather than AI tells and must be preserved unconditionally, even in --mode all.

### Unconditional Preservation List

| Pattern | Why it's voice, not AI |
|---|---|
| Lowercase sentence starts (`closed our seed on a tuesday...`) | Deliberate cadence cue. |
| `..` as a soft pause | The officially-blessed alternative to em dash. |
| Sentence fragments (`Worth it.`, `Every time.`) | Pass 2 adds fragments; existing ones must stay. |
| Contractions (`don't`, `it's`) | Mandatory for natural rhythm. |
| First-person sensory detail (`my hands shook`) | Required for Pass 3. |
| Specific numbers (`$47k`, `9:14am`) | Required for Pass 3. |
| Named entities (`HubSpot`, `Tuesday morning`) | Must be capitalized properly. |
| Self-correction (`actually no`, `correction:`) | Burstiness signal. |

The humanizer uses a three-pass system to process text:

Pass 1: SCRUB (Delete or Replace)

Uses tiered regex catalogs to clean text:

• FORENSIC (Always on): Targets model leakage like oaicite, contentReference, turn0search0, and knowledge-cutoff disclaimers.
• STRICT (Default): Targets corporate-speak and bad LinkedIn style. Includes vocabulary swaps (e.g., leverage $\rightarrow$ use), filler adverb deletion, and banning negative parallelism.
• AESTHETIC (Opt-in): Targets patterns AI uses that humans also use (e.g., rule-of-three, "robust", em dashes). Use with caution as it can flatten good writing.

Pass 2: BREAK (Force Burstiness)

Targets readability metrics (Flesch reading ease >55, sentence length variance >40%) by:

• Breaking long sentences into fragments.
• Adding asymmetric sentence structures.
• Varying cadence.

Pass 3: ADD (Human Fingerprints)

Injects human-like elements (without fabricating facts):

• Specific numbers (replacing "many" or "significant").
• Named entities (people, companies, dates).
• First-person sensory details or moments of vulnerability.

The linkedin-humanizer package uses a three-tier taxonomy to categorize patterns flagged as 'AI-like'. This taxonomy helps users decide whether to accept a rewrite, defend a stylistic choice, or audit a detector's false positive.

The Three Tiers

1. Forensic (Undefendable): Real leakage from models or prompt templates. These are accidental artifacts that no human produces.

   • Examples: oaicite markers, contentReference tokens, turn0search0 artifacts, knowledge-cutoff disclaimers, or [Your Name] placeholders.
   • Defense Strength: Zero.

2. Strict (Corporate-speak): Real human patterns that are often banned by users to maintain a specific brand voice. They don't prove AI usage, but they are considered 'cheap' or cliché.

   • Examples: Words like leverage, utilize, harness, delve, foster, cultivate; filler adverbs (fundamentally, essentially); or cliché openers like "In today's fast-paced world."
   • Defense Strength: Medium (can be defended as professional vocabulary, but usually overridden by brand-voice requirements).

3. Aesthetic (Overreach/Defendable): Patterns flagged because LLMs happen to use them, even though they are hallmarks of famous human writers. These often cause false positives in detectors.

   • Examples: Em dashes (Emily Dickinson), the rule of three (Lincoln), passive voice (scientific writing), or the word "robust" (engineering/epidemiology).
   • Defense Strength: High.

The linkedin-humanizer uses a three-tier classification system to identify and manage signals that AI detectors (like OriginalityAI or GPTZero) use to flag content as AI-generated. Understanding these tiers helps in deciding which writing patterns to modify and which to defend.

Tier 1: Forensic (Real AI Signals)

These are definitive markers of LLM generation. They have zero defense strength because they are technical artifacts of how AI models function or output data.

• Examples: oaicite / contentReference / turn0search0 markers, knowledge-cutoff disclaimers, unfilled phrasal templates (e.g., [Your Name]), Mad-Libs style blanks, and extreme em-dash overuse (3+ in a single short post).

Tier 2: Strict (Corporate-speak, Easy Ban)

These are stylistic patterns common in LLM outputs that are frequently used in corporate or influencer writing. They have low to medium defense strength.

• Examples: Specific AI vocabulary (e.g., leverage, utilize, harness, delve), filler adverbs (fundamentally, essentially), and cliché openers ("In today's fast-paced world").

Tier 3: Aesthetic (Overreach, Defendable)

These are standard human rhetorical or typographic patterns that AI detectors often incorrectly flag. They have high defense strength because they are foundational to human literature, science, and professional tools.

• Examples: Single em-dash use, the 'Rule of Three', passive voice in technical contexts, specific technical terms like robust, and curly quotes (smart quotes).