Start with a marketing job, give the agent clear rules, then reuse it whenever you need the same answer again.
Add the OpenRush skill
A skill teaches your AI assistant when and how to use OpenRush. Add this line to your assistant’s instructions:Good agent jobs
Start with one job. Keep it narrow enough that you can tell whether the answer is useful.Competitor scan
Keyword gap review
Opportunity list
SERP snapshot
Website audit
Weekly growth brief
Write the agent instructions
Use a short brief. Good agents do not need long prompts.Use Claude Code
Claude Code supports custom subagents. Use a subagent when you want a reusable specialist, such as a growth analyst, SEO reviewer, or content brief writer. To create a subagent:- Connect OpenRush with MCP from Setup.
- In Claude Code, run
/agents. - Create a personal or project agent.
- Give it a clear description, such as
Use this agent for organic search research with OpenRush. - Limit tools where possible. A research agent usually only needs read and connector access.
- Test it with one site before using it repeatedly.
- Every Monday, create a competitor brief for your domain.
- Every Friday, find new keyword gaps to review.
- After a launch, check priority keywords and search results.
- Monthly, audit the site and list the highest-impact fixes.
/schedule in Claude Code when it is available for your account.
Claude Code routines are in research preview. Review the first few runs before you rely on them.
Use Codex
In Codex, make the workflow durable with three pieces:- MCP: connect OpenRush so Codex can call the data tools.
- AGENTS.md: save your working rules, voice, output format, and review checklist.
- Automations: schedule recurring checks when the task is stable.
AGENTS.md section can be enough:
Schedule useful marketing tasks
Scheduled tasks work best when the job has a clear cadence and a clear output. Use Claude Code routines or Codex automations depending on where you already work.Prompt templates
Weekly competitor brief
Keyword gap queue
Website audit
Best practices
- Start with one narrow job.
- Test manually before scheduling.
- Keep the output format the same every time.
- Ask for a short answer first, then drill into details.
- Review the first few scheduled runs before trusting the workflow.
- Keep credit usage in mind. Recurring jobs use the same OpenRush account credits as normal MCP and API calls.
OpenRush provides the marketing data. Your AI client controls agent setup, memory, scheduling, and permissions.