Choosing Between a Skill and a CLAUDE.md Entry

A backend team wants Claude Code to reliably know two things:

• Fact A: "We use uv instead of pip for all Python dependency management in this repo."
• Fact B: A twelve-step incident-response runbook (who to page, which dashboards to check, in what order, with the exact kubectl commands for each step) that's needed maybe once a month, when there's an active production incident.

1. Recommend where each fact should live, and justify the choice using the cost model of each mechanism.
2. For Fact B specifically, would you set disable-model-invocation: true? Justify your answer.
3. Suppose the team instead put both facts in CLAUDE.md. What concrete problem would this cause as the runbook grows over time?