What Goes in an Agent Skill for Social Media Posting
Facts checked on October 13, 2026
An agent skill is a folder with a file called SKILL.md that tells an AI assistant how to do one job. A skill for social media posting should hold five things: your brand voice, each network’s length limit, when to ask you for approval, which tool publishes, and what never to post. The skill is instructions only. To publish, the assistant still needs a tool that is connected to your social accounts, such as a connector in the chat app (MCP), a command-line tool or an API.
What an agent skill is#
The Agent Skills specification defines it in one line: “A skill is a directory containing, at minimum, a SKILL.md file”. In plain words, a folder with one text file in it. The format’s site says it “was originally developed by Anthropic, released as an open standard”.
The file has two parts: a short header at the top with the skill’s name and description, then the instructions in ordinary text.
| Part | What the specification asks |
|---|---|
| Name | Required. Up to 64 characters: lowercase letters, numbers and hyphens, the same as the folder’s name |
| Description | Required. Up to 1,024 characters. “Describes what the skill does and when to use it.” |
| Instructions | “There are no format restrictions. Write whatever helps agents perform the task effectively.” |
The description matters most. An assistant first reads only the name and description of each skill, and opens the instructions when a request matches. So write the description in the words you would use to ask: “draft a LinkedIn post”, “schedule a thread”.
Keep it short. The specification says: “Keep your main SKILL.md under 500 lines.”
Which assistants read skills#
Each row is what that assistant’s own documentation says.
| Assistant | What its documentation says |
|---|---|
| Claude Code | “Claude Code skills follow the Agent Skills open standard” |
| Claude, the chat apps | “Skills are available for users on Free, Pro, Max, Team, and Enterprise plans.” The code execution setting has to be on. You upload the folder as a ZIP file |
| ChatGPT and Codex | “Skills build on the open agent skills standard.” Standalone skills work in the ChatGPT desktop app and in Codex |
| Cursor | “Skills work across any agent that supports the Agent Skills standard.” |
| Gemini CLI | Agent Skills “let you extend Gemini CLI with specialized expertise”. You are asked to confirm before a skill is used |
| OpenClaw | “OpenClaw follows the AgentSkills spec.” |
One folder can serve several of them, because they read the same format. Where the folder goes differs by assistant.
What a social posting skill should contain#
Brand voice#
Say who is speaking and to whom. Give two or three real posts as examples, and the words you avoid. An assistant copies an example more closely than it follows an adjective.
Length limits per network#
A post that is too long is refused or cut. Put the limits in the skill so the assistant writes one version per network. These are the figures Postnavo’s site states for its character counter. They were not read again from each network for this post.
| Network | Limit on Postnavo’s site |
|---|---|
| X | 280 characters for standard accounts. A link counts as 23 |
| Bluesky | 300 characters per post |
| Threads | 500 characters per post |
| 2,200 characters for a caption | |
| 3,000 characters for a post |
When to ask for approval#
Write the rule as a step: show the full text, the channels and the time, then wait for a yes. Say which cases always need a person, such as a reply to a customer or anything with a price.
Which tool or command publishes#
Name the one way out: the exact command, or the connector’s tool. Say what to do when it fails: report the problem and stop, never keep retrying.
What never to post#
A short list is enough: customer names, unreleased prices, legal or medical claims, replies to a complaint, anything about a competitor.
An example SKILL.md skeleton#
This is an illustration written for this post. It is not a file Postnavo ships, and the publishing command is a placeholder. The lines between the two --- marks are the header; the rest is the instructions.
---
name: social-posting
description: Draft and schedule social posts in our brand voice. Use when asked to write, schedule or announce on X, LinkedIn, Threads or Bluesky.
---
## Voice
Plain and friendly. No hype words. Examples: references/posts.md
## Limits
X 280. Bluesky 300. Threads 500. LinkedIn 3,000. One version per network.
## Approval
Show the text, channels and time. Wait for "yes".
## Publish
npx postnavo-cli post --to x,linkedin --at "<time>" "<text>"
If it fails, report the error and stop.
## Never post
Customer names. Unreleased prices. Replies to complaints.What a skill cannot do by itself#
It cannot publish. A skill is text. The format’s overview calls it “instructions that tell an agent how to perform a specific task”, and OpenClaw’s documentation calls skills “markdown instruction files that teach the agent how and when to use tools”. No page read for this post describes a skill that can sign in to a social network itself. The post goes out through a tool that is connected to your accounts:
- a connector in the assistant (an MCP server). See how to post to social media from Claude with MCP;
- a command-line tool the assistant runs on your computer;
- an API that a script or an automation tool calls.
For how the three compare, see what an agentic social media scheduler is.
It may not reach the internet at all. Anthropic’s Agent Skills overview gives a different answer per product. Through Claude’s developer platform, skills cannot reach the internet. In the chat apps, internet access is full, partial or none, depending on settings. In Claude Code, skills have “the same network access as any other program on the user’s computer.”
It cannot enforce its own rules. The specification describes the body as instructions. It describes nothing that makes an assistant follow them. Treat “wait for a yes” in a skill as a habit you teach, and keep a real approval step in the assistant’s settings or in the scheduler.
It is not safe just because it is short. Anthropic’s overview says: “Use Skills only from trusted sources”. Read a skill before you install it.
What Postnavo is built to do#
Postnavo is built to be the tool on the other side of the skill. It is launching with a command-line tool and an installable agent skill: a small instruction package that tells a coding agent which command to run, and when. Where it installs for each agent will be published at launch. See the Claude Code page.
For the chat apps, Postnavo is launching with a hosted MCP connector that you link by signing in. In Claude’s chat apps that connector is the route, not the skill. Every assistant Postnavo is launching with is on one page.
Postnavo is also launching with an approval setting of its own. While it is on, a post an agent creates will wait for your OK, whatever the skill said. Your brand voice and your never-post list stay yours to write, in a skill of your own.
FAQ#
Is a Claude skill the same as an MCP server?#
No. A skill is a folder of instructions. An MCP server is a connector: a program that gives the assistant actions it can take, such as scheduling a post. A posting setup usually needs both: the skill says how to write and when to ask, and the connector or the command publishes.
Can I use one skill in Claude Code and Codex?#
The format is shared: both products’ documentation names the Agent Skills standard. The folder each one reads from is different, and Claude Code adds settings of its own that other assistants may ignore.
Does a skill need code?#
No. The specification requires only the SKILL.md file. Scripts and reference files are optional.