# Why your Claude writing skill doesn't sound like you > Most free Claude writing skills remove AI tells rather than learn your voice. Here is the difference, and why a skill cannot post on your behalf. URL: https://voicemoat.com/blog/claude-writing-skill-voice ## Key facts - Published: 2026-09-09 - Category: AI and Voice - Read time: 13 minutes - Author: Prateek Singh - Publisher: VoiceMoat - Topics: Claude Skills; SKILL.md; Writing voice; Model Context Protocol; AI writing detection ## Key points - Almost every free writing skill is subtractive. It removes phrasing that reads as machine-written, measured against a general standard, not against you. The best known has over 45,000 stars and a dozen translated forks. - Removing AI tells does not produce your voice. It produces competent anonymous prose. Those are different problems and only one of them is being solved. - Skills that genuinely learn a voice from your own writing do exist, and they sit at 264, 40, 18 and 9 stars. Two orders of magnitude behind the humanizers. - The largest free marketing collection, at nearly 49,000 stars, has no voice-deriving skill at all. Its copywriting skills consume a voice you supply and never build one. - A skill cannot post anything. Not because of a missing permission, but because there is no credential, no session and no authentication anywhere in the format. That absence is the design, not an oversight. - Star counts mislead badly here. The two most-starred skill collections on GitHub contain zero skills between them. They are lists of links. ## Related articles - Can Claude post to LinkedIn? Yes, and LinkedIn's own developer terms say it should not: https://voicemoat.com/blog/can-claude-post-to-linkedin - Which LinkedIn MCP servers stop working after 60 days: https://voicemoat.com/blog/linkedin-mcp-60-day-token - Does LinkedIn have an analytics API? Yes, and almost no guide mentions it: https://voicemoat.com/blog/linkedin-analytics-api ## FAQ Q: Can a Claude skill post to LinkedIn or Twitter? A: No. On the Claude API, skills run in a sandbox with no network access. Even in Claude Code, where a skill has full network access, there is no credential, no session and no authentication mechanism in the skill format at all. Publishing requires an MCP connector, which is a different thing entirely. Q: Why doesn't my Claude writing skill sound like me? A: Most likely because it was never trying to. Almost every free writing skill is subtractive: it removes phrasing that reads as machine-written, measured against a general standard rather than against your writing. That produces prose that reads as human and anonymous, which is a different outcome from sounding like you. Q: Where do Claude skill files go? A: In Claude Code, a folder per skill inside a skills directory, either in your home .claude directory for personal use or the project's .claude directory. On claude.ai and Claude Desktop there is no filesystem path: you upload a ZIP under Customize then Skills, with the skill folder at the root of the ZIP rather than nested. Q: What is the difference between a Claude skill and an MCP server? A: A skill is a folder of instructions that teaches Claude how to do something well. An MCP server is a running service that lets Claude reach an external system. A skill needs no server and holds no credentials. A connector needs both. Anthropic's own guidance to builders is to ship a connector first and then skills that help people use it. Q: Why does uploading a skill need code execution enabled? A: Because of how a skill reaches Claude. When a skill matches your request, Claude opens the file and reads it using a shell command. Without a filesystem and a shell there is no way to open it. The setting is not there so your skill can run code, it is there so Claude can read your markdown. Q: Are Claude skills safe to install? A: A skill is instructions Claude will follow, so treat it like installing software rather than reading an article. One February 2026 study of 3,984 skills from two community registries found roughly 37 percent with a security issue of any severity, 13 percent with a critical one, and 76 confirmed malicious after review. Prose-only writing skills are the low-risk end; skills bundling scripts or fetching URLs are not. Q: How long can a skill description be? A: Anthropic's own documentation disagrees with itself. The claude.ai help page says 200 characters maximum, the platform documentation says 1,024, and both were live in September 2026. If you intend to upload to claude.ai, stay under 200 until a real upload proves otherwise. Source: https://voicemoat.com/blog/claude-writing-skill-voice Site overview: https://voicemoat.com/llms.txt. AI usage policy: https://voicemoat.com/ai.txt.