Social media agent skill builder

Pick what your agent should do and how much it may do on its own, then copy or download a SKILL.md, AGENTS.md or system prompt. The file uses PostBoost's real MCP tool names and keeps the agent on drafts, approval and a daily cap.

File
How much may the agent do on its own?
Connection
Tasks
Network rules (optional)

One rule per line. The first 10 lines are used, up to 200 characters each.

SKILL.md


    

Install

    Runs in your browser. Nothing you type is sent to PostBoost.

    Use this tool from code

    The same tool is a free API call, no account needed. It also works with our 9 SDKs and the MCP server.

    curl
    curl -X POST https://postboost.co/app/api/tools/social-media-agent-skill-builder -H 'Content-Type: application/json' -d '{"brand":"Acme Bakery","tasks":["draft_posts"],"networks":["instagram"],"daily_cap":3}'

    API docs for this tool SDK quickstart

    Last verified:

    Sources:

    A skill that drafts first

    Most social media skills give an agent every tool and hope for the best. This builder starts from the other end: by default the agent may only save drafts, and a person schedules them in PostBoost. You can let it schedule after your approval or add posts to the queue on its own, and the file says exactly which tools it may call at each level.

    What goes into the file

    • Setup: how to connect the agent to PostBoost, through the connector (OAuth) or the advanced server (API key).
    • Guardrails: your autonomy level, a daily cap, validation before every write, no permanent deletes, no repeat posts unless you ask, and an idempotency key so a retry never creates a second post.
    • Workflows: one numbered list per task, naming the PostBoost MCP tools in the order the agent should call them.
    • Network rules: the caption limit and notes for each network you pick.
    • Your rules and example prompts: your own do's and don'ts, and two prompts per task to try first.

    Facts

    • Agent Skills name fields are 1 to 64 lowercase letters, digits and hyphens, with no hyphen at the start, at the end or twice in a row, and match the skill's folder; description is at most 1,024 characters; the specification recommends keeping SKILL.md under 500 lines (agentskills.io, checked 2026-10-21).
    • AGENTS.md is plain Markdown with no required fields (agents.md, checked 2026-10-21).
    • The builder takes up to 10 rules of 200 characters each. Its largest file stays under 500 lines.

    How to use

    1. Pick the file: SKILL.md for Claude and other skill-aware agents, AGENTS.md for coding agents, or a system prompt for ChatGPT and other assistants.
    2. Tick the tasks the agent should handle, such as drafting, scheduling, reports or fixing failed posts.
    3. Choose how much the agent may do on its own: drafts only, ask before scheduling, or queue without asking. Set a daily cap.
    4. Add the networks you post to and your own rules, such as words to avoid.
    5. Copy or download the file, save it where the install steps say, and connect your agent to PostBoost.

    FAQ

    What is an agent skill?

    An Agent Skill is a folder with a SKILL.md file: a short YAML header with a name and a description, then plain instructions. The agent reads the name and description at startup and loads the instructions when a request matches, so a social media skill only enters the conversation when you ask about posts.

    What is the difference between SKILL.md, AGENTS.md and a system prompt?

    SKILL.md follows the Agent Skills format and loads on demand in agents that support skills, such as Claude and Claude Code. AGENTS.md is a plain Markdown file in a project's root that coding agents such as Codex and Cursor read for every task. A system prompt is the same instructions pasted into an assistant's custom instructions, for example in ChatGPT. The builder writes the same rules in all three.

    Will my agent publish without asking?

    That depends on the setting you pick. Drafts only: the agent saves drafts and a person schedules them in PostBoost. Ask, then schedule: the agent shows you each post and schedules it only after your yes. Queue without asking: the agent may add posts to your queue on its own, so use it only where every post may go out unreviewed. Every setting tells the agent to validate posts first, never delete posts permanently and stop when PostBoost refuses a write.

    Does it work with ChatGPT, Cursor and Codex?

    Yes. Use a system prompt in ChatGPT, AGENTS.md or a system prompt in Cursor and Codex, and SKILL.md in Claude and Claude Code. The agent also needs the PostBoost MCP server: the connector with OAuth sign-in, or the advanced server with an API key. The MCP config generator shows the setup for each client.

    Does this page store what I type?

    No. The file is built in your browser and nothing you type is sent to PostBoost or saved. The builder uses no AI and costs no credits. The same builder is available through the free tools API and MCP server if you want to generate files from a script.

    Why does the API key option ask for a scoped key?

    With drafts only and an API key, the file tells you to create a key with only the read and write abilities and without publish. The advanced server then refuses to publish with that key, even if the agent tries. With the connector, which signs in with OAuth, the guardrails in the file are what keep the agent on drafts.

    Can I edit the file after I download it?

    Yes. It is plain Markdown. Keep the name lowercase with hyphens and the same as its folder, keep the description under 1,024 characters, and keep the file under 500 lines, as the Agent Skills format asks.

    Related

    Try PostBoost free

    API, webhooks and MCP on every plan, including Free. No credit card required.