content-planner is an editorial planning workflow for AI agents. It pulls 90 days of Search Console data for a property, sorts every query and page pair into opportunity buckets, scores each by click potential and schedules the best ones as a calendar.
The result is a Markdown calendar with optional JSON, capped at 12 topics for a three-month plan. It refuses to plan from too little data and flags cannibalization before anything new is scheduled.
What You Get
- A dated calendar of up to 12 topics, one post per week, highest priority first
- Each topic tagged as striking-distance, gap, CTR underperformer or related expansion
- A click-potential score and a one-sentence rationale per topic
- A writer prompt per topic to paste into a content writer skill
- Warnings for cannibalization and missing business context
Common Use Cases
- Deciding what to write next from real search data
- Finding pages stuck at positions 5 to 20 that a refresh could lift
- Spotting pages with good rankings but weak click-through rates
- Merging new topics into an existing calendar without losing manual edits
What Makes It Unique
It plans from the site’s own Search Console rows and not from brainstormed topics, which keeps keyword discovery with keyword-research and writing with content-writer. A quality gate stops the plan if there is no live data or fewer than 50 query and page rows, and rejects bare-keyword titles and duplicate primary keywords. Existing in-progress or published entries are preserved when it merges with a calendar the user supplies.
Under The Hood
The skill is a short wrapper whose SKILL.md tells the agent to read a longer WORKFLOW.md in the seo folder of the same repository and follow it. The workflow is prose only and ships no scripts. It reads Search Console data through the connected capability described in the analytics workflow, or from a supplied export, and writes the calendar only to the destination the user chooses. It starts no viewer or publishing service, never publishes to a CMS without an explicit yes, and treats fetched data as untrusted.