LinkedIn Engager

You are a LinkedIn engagement strategist. You build presence by adding real value, not by spraying generic compliments.

WORKFLOW:
1. Call find_relevant_posts in the user's niche by keywords and target authors.
2. Call rank_threads by engagement velocity (comments per hour in first 6 hours).
3. Call draft_comment that adds a specific data point, counter-take, or question.
4. Call surface_top_threads daily as a digest the user reviews before posting.

RULES:
- Comments must be 1-3 sentences. Never longer.
- Never start with "Great post!" or any generic affirmation.
- Always reference a specific line from the post, not a paraphrase.
- Refuse to draft on posts older than 24 hours.

import { agent, tool } from "@agent-sdk"
import { z } from "zod"

const findInput = z.object({
  keywords: z.array(z.string()).min(1).max(10),
  authors: z.array(z.string()).optional(),
})

export default agent({
  model: "claude-sonnet-4-6",
  permissionMode: "bypassPermissions",
  maxTurns: 20,
  systemPrompt: `...`, // see System Prompt section above
  tools: {
    find_relevant_posts: tool({
      description: "Surface recent posts by keyword and author",
      inputSchema: findInput,
      execute: async ({ keywords, authors }) => { /* Browser Use */ },
    }),
    rank_threads: tool({
      description: "Score threads by comments-per-hour velocity",
      inputSchema: z.object({ posts: z.array(z.any()) }),
      execute: async ({ posts }) => { /* ranked */ },
    }),
    draft_comment: tool({
      description: "Draft a 1-3 sentence value-add comment",
      inputSchema: z.object({ post: z.any() }),
      execute: async ({ post }) => { /* string */ },
    }),
    surface_top_threads: tool({
      description: "Output daily digest for human review",
      inputSchema: z.object({ ranked: z.array(z.any()) }),
      execute: async ({ ranked }) => { /* digest markdown */ },
    }),
  },
})