← All use cases

中文

Sales & marketing · Claude Code

Set up a lead marketing agent that hands tasks to specialist subagents

Idea + starter prompt, not run by us yet

For: Marketers who run many small recurring tasks across channels

A marketing manager agent in Claude Code reads the weekly plan, checks which three-hour slot it is, and starts specialist subagents stored as markdown files in .claude/agents/. They use custom scripts for Reddit, Hacker News, Gemini images, a CDN and Slack. It automated recurring work, but in week one the agent handling social posts published to Reddit twice, because the rule against duplicates was too vague.

Steps

  1. Write your weekly plan as a simple file with time slots and tasks.
  2. Create one subagent per job, each with a markdown file that states its tools and limits.
  3. Make rules specific. A vague 'no duplicates' did not stop repeated Reddit posts, so say exactly how to check what was already posted.
  4. Run every channel in draft or approval mode first.
  5. Check the logs daily in the first weeks and tighten the rules where it slips.

Starter prompt

Starter prompt. Change the [brackets] first.

Trial first: run one subagent for one time slot, using a dummy [log file] and a made-up topic. Write its draft to a folder called [trial-drafts]. Do not connect any real channel.

I want a marketing manager agent and three specialist subagents: [social, content, reporting], set up as markdown files in .claude/agents/. The manager reads [weekly-plan.md], checks which time slot it is and hands the matching task to the right subagent. For each subagent write exact rules, including how to check [a log file] for what was already posted, and never the same text twice. Show me the files first.

In this first version every subagent only writes drafts to a folder. Stop and wait for my yes before anything is posted to any channel. A draft is not permission. Do not ask for passwords.

Original post: Reddit · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.