drafting-a-skill

Pipeline-only content, never invoked via the Skill tool -- its own procedure text is read and run directly by executing-a-branch-plan (Step 6, agentType branch-plan-task) for a brand-new or existing SKILL.md, or by scorer-gated-skill-edits's own Step 3 for one bounded gate-loop iteration or Step 9 for its pre-ship review -- never the entry point for "should this even be a skill."

Compatibility

disable-model-invocation is a Claude-Code/Cursor-specific field the standard Agent Skills spec does not define; on a runtime without it, this skill's manual-only invocation mode must be enforced by naming convention or documentation instead.