marketing-loops helps an agent turn a marketing task you would forget into a repeatable loop: a weekly SEO scan, an ad-fatigue check, a churn watch. Every loop has nine parts, including when it acts, what it remembers between runs and when it stops.
References cover a 61-loop catalog, guardrails for spending, sending and publishing, state files, a template, and two bundled operators for SEO and outbound.
What You Get
- A loop chosen or adapted from a catalog of 61 loops across SEO, paid, outbound, activation, retention and referral work
- A nine-part loop spec: cadence, action condition, purpose, skills used, steps, self-check, state, stop rule and output
- A cadence matched to how fast each signal changes
- A two-tier action model with spend, send and publish guardrails and a kill switch
- An SEO operator and an outbound operator that bundle loops with shared memory and weekly reports
Common Use Cases
- Setting up a weekly SEO opportunity scan for a marketing site
- Watching ad creative fatigue every few days
- Triaging outbound replies and sending-domain health
- Flagging accounts that cross a churn risk threshold
- Writing a new loop from a template when the catalog has none
What Makes It Unique
Every loop must state its stop condition, a self-check and what it remembers, and the skill separates how often a loop looks from when it acts. It warns against over-frequent and vanity loops, ranks actions into autonomous-safe and gated tiers, and bundles whole SEO and outbound jobs with shared memory.
Under The Hood
The SKILL.md and its references are loaded as context, and the large loop catalog is read when a loop is chosen. It ships no scripts. The agent reads the product marketing context file in your project. Loops it sets up keep state in JSON and log files under an agents folder in your project, and the SEO and outbound operators keep working files there too, so the agent writes files. It suggests scheduling through Claude Code’s loop and cron features or any cron job, and its loops may read analytics and Search Console data, crawl a site, draft emails and open pull requests. Spending, sending and publishing are staged for human approval by default.