Ship fixes with your coding agent

GetIntel finds what is missing; your coding agent can change the real site. This playbook walks through one fix end to end, in Claude Code. Cursor and Codex CLI work the same way.

You need: GetIntel connected over MCP with a token that has action access (see Connect over MCP), and your agent open in your website’s repository.

1. Find the work. Ask:

   What are my top three On-Page tasks this week? Show why each matters.

   The agent calls get_weekly_plan or list_tasks and summarizes them.

2. Read the brief. Ask:

   Open the first task and tell me exactly what needs to change.

   The agent calls get_task, which returns why it matters, the steps, and the topics it moves.

3. Pull the context. For a content task, ask:

   Get the topic this task moves and list the facts AI compares that our page doesn’t cover.

   The agent calls get_topic and reads its content analysis.

4. Make the change. Ask:

   Find the page in this repo and update it to cover those facts, in our existing style. Don’t invent numbers; mark anything you’re unsure about.

5. Review and ship. Read the diff. When it’s right:

   Commit this on a new branch and open a pull request.

6. Update GetIntel. Ask:

   Mark the task as done.

   The agent calls update_task.

Caution

Review every change before it merges. The agent works from GetIntel’s evidence, but you know your product: check facts, prices and claims.

Technical fixes

For technical tasks, the loop is shorter and verifiable:

Check the latest technical scan, fix the robots.txt and canonical issues in this repo, then re-run the technical scan.

The agent reads get_technical, makes the fixes, and after you deploy, calls recheck_technical (once an hour). Results appear in get_technical a few minutes later.

Measure

Content changes usually take a few weeks to a month to show in topic scores. Ask your agent later: “How has the topic this task moved changed since we shipped the fix?”