Linkedin Humanizer

Aggressively rewrites LinkedIn text to remove AI indicators and add human traits, ensuring posts and comments read authentically before publishing.

Security Scan

The skill's stated purpose (rewriting LinkedIn copy to remove AI tells) matches its code-free instructions, but there are internal contradictions and behaviour that could enable deceptive publishing — worth caution and clarification before installing.

Purpose & Capability

The name/description match the SKILL.md and the provided regex/rules. It is an instruction-only text-rewrite skill that uses regex replacements, sentence restructuring, and required 'human fingerprints' — all coherent with 'humanizer' functionality. No unrelated binaries, env vars, or install steps are requested.

Instruction Scope

Overall the instructions stay inside rewriting scope (regex scrubbing, structural edits, and asking the user when data is missing). However there are notable contradictions and risky behaviors: the doc mandates 'Never introduce facts that weren't in the input' and 'Don't fabricate' yet the example output adds a concrete number and named entity not present in the input (35k, LinkedIn) — this inconsistency raises the risk the agent might fabricate despite the rule. Also the 'ADD' pass requires adding a named entity and a specific number per 100 words; if the user doesn't provide them the instructions say to ask, but the requirement itself could pressure adding facts that change perceived truthfulness. The SKILL.md also promises a 'per-sentence perplexity estimate' but gives no algorithm or safe guardrails for how that metric is computed or whether it requires external model calls.

Install Mechanism

No install spec and no code files beyond instruction and reference docs. Instruction-only skills have low install risk because nothing is written to disk or downloaded automatically.

Credentials

The skill requests no environment variables, no credentials, and no config paths. There is no request for unrelated secrets or system access.

Persistence & Privilege

The skill is not marked always:true and is user-invocable only; it does not request to modify other skills or system settings and has no install-time persistence.

What to consider before installing

This skill is coherent with its stated task (rewriting LinkedIn text to hide AI patterns) but has two issues you should consider before installing: - Contradictory rules about fabrication: The doc explicitly forbids inventing facts and says to ask the user for missing numbers/anecdotes, but the example shows fabricated facts. Ask the author to confirm and to enforce a human-in-the-loop: require explicit user consent before inserting any numbers, names, dates, or other facts. Prefer a default behavior of refusing to add facts without user-provided values. - Potential for deceptive use: The tool is designed to evade AI detectors. That makes it useful for legitimate editing, but also for deceptive publishing or policy violations. Consider whether using such a tool could violate LinkedIn's terms, company policy, or your own ethical boundaries. Practical steps before use: - Request clarification from the skill author about the apparent example fabrication and ask them to update SKILL.md so the behavior is unambiguous (e.g., require interactive prompts for any missing specifics). - Test thoroughly with non-sensitive drafts and confirm the skill never injects new factual claims without explicit user approval. Verify the diff output is clear and shown to the user before finalizing. - Prefer workflows that keep a human review step and log changes. If you need to comply with workplace rules, do not use this skill to alter truth claims or publish content that could mislead others. Given these contradictions and the potential for misuse, treat the skill with caution and get the author to remove the ambiguity around fabrication and required 'human fingerprints' before enabling it broadly.

Aggressively rewrites any text to pass AI detectors and read authentically human. Based on Wikipedia's "Signs of AI writing" taxonomy plus 2026 LinkedIn-specific patterns.

When to use

• Before publishing any AI-drafted post or comment
• When linkedin-post-audit flags AI tells
• When a draft feels "off" and you can't pinpoint why

Input

Any text (post, comment, reply, DM). Optional: target voice samples (past human posts by the user).

Output

• Rewritten text with AI tells removed
• Diff showing what changed and why
• Per-sentence perplexity estimate (higher = more human)
• Confidence: "human", "mixed", "AI-likely"

The three passes

Pass 1 — SCRUB (delete or replace)

Punctuation:

• — → . or ,
• – → - or to
• -- → . or ,
• " " → "

Vocabulary (regex-strip and replace):

• leverage → use
• utilize → use
• facilitate → help
• streamline → simplify
• delve → look
• navigate → handle
• unlock → find
• harness → use
• foster → build
• cultivate → grow
• fundamentally → (delete)
• essentially → (delete)
• ultimately → (delete)
• crucially → (delete)
• notably → (delete)
• landscape → field (or delete)
• ecosystem → (contextual)
• paradigm → approach
• realm → area
• robust → solid
• seamless → smooth

Phrase-level:

• "It's not just X, it's Y" → rewrite as a single claim
• "In today's fast-paced world" → delete opener entirely
• "game-changer" → specific descriptor
• "deep dive" → "look" or "analysis"
• "at the end of the day" → delete

Pass 2 — BREAK (force burstiness)

Target: Flesch reading ease >55. Sentence length variance >40%.

• If all sentences are 15-22 words, force-break at least 1 in 3 into <8-word sentences
• Add at least one sentence fragment ("Worth it.", "Every time.")
• Break rule-of-three lists into twos or fours
• Break perfect parallel structures with one asymmetric sentence

Pass 3 — ADD (human fingerprints)

Require at least:

• 1 specific number per 100 words (replace "many" / "significant" / "massive")
• 1 named entity (real person, company, date, city)
• 1 first-person sensory detail
• 1 contradiction or self-correction
• 1 moment of vulnerability or stakes

If the input lacks these, ask the user for a specific number or anecdote to plug in. Don't fabricate.

Non-negotiable rules

• Preserve the user's actual claim. Humanizing ≠ changing meaning.
• Capitalize all names (Dharmesh, Felix, HubSpot, Claude).
• Never introduce facts that weren't in the input. If a number is missing, ask.
• Keep the user's sentence-level voice quirks (lowercase starts, .. soft pauses).

Example

Input: "In today's fast-paced landscape, businesses must fundamentally leverage AI to unlock robust ROI — here's what I've learned."

Output: "businesses need AI to cut costs. here's what we learned running 35k LinkedIn profiles through our system daily."

Diff: removed em dash, removed "in today's fast-paced landscape", removed "fundamentally", removed "leverage", removed "unlock", removed "robust", added specific number (35k), added named entity (LinkedIn).

Files

• SKILL.md — this file
• references/scrub-rules.md — full regex patterns and replacement mapping
• references/voice-fingerprint.md — how to preserve user voice while scrubbing

Related skills

• linkedin-post-audit — detection-only pass (no rewrite)
• linkedin-post-writer — generates drafts that already pass the humanizer