Use onWatch when an operator needs a single preflight and in-run view of quota burn across coding-agent providers, not when they are simply using one provider dashboard normally. The invoke moment is concrete: before or during a long coding-agent run, poll provider usage endpoints, store history, surface remaining headroom, and alert before throttling or overage hits. That scope boundary, cross-provider quota and usage monitoring for coding-agent operations, makes this publishable as a skill instead of a generic LLM product or dashboard listing.