Agent skill

The MCP tools give your agent hands; the skill gives it manners. It's one markdown file that teaches Claude the workflow that makes xpost work well: list accounts first, dry-run guardrails before posting, treat pending_approval as success, read the per-row bulk report, and report delivery honestly — including partial failures.

Install for Claude Code

mkdir -p ~/.claude/skills/xpost
curl -s https://dev.xpost.to/docs/dist/SKILL.md \
  -o ~/.claude/skills/xpost/SKILL.md

Project-local works too: put it in .claude/skills/xpost/ inside the repo your agent works from. Claude picks it up automatically whenever a task smells like social posting.

What it teaches

• · The full workflow: accounts → media → post → receipt
• · Per-platform configuration — caption overrides, stories/reels placement, Pinterest boards, X first comments and threads
• · Approval etiquette: a held post means "tell the user", a guardrail block means "rewrite once, never evade"
• · Bulk posting with the per-row report, and the curl fallback when MCP isn't available

Read it first

The whole skill is public: SKILL.md ↗. The local stdio MCP server is downloadable too (server.mjs + tools.mjs, needs npm i @modelcontextprotocol/sdk zod) — though the hosted connector needs no download at all.