claude-power-tools

Proactively spot moments in ordinary work where one of Claude Code's force-multipliers should be proposed - a multi-agent Workflow (fan-out audits, repo-wide sweeps, judge panels, adversarial verification), an agent team (several separable long-lived tracks that must coordinate an evolving contract and that you want to steer in parallel - e.g. a cross-layer feature spanning frontend, backend, and tests), a /goal (a mission lacking verifiable success criteria), a /loop or /schedule (babysitting, polling, recurring or deferred obligations), capturing a just-invented repeatable technique as a skill, a settings hook for "I keep manually doing X after Y" patterns, recording a durable design principle in a decision log, or recalibrating effort (cheap-vs-heavy models, quick-pass-vs-adversarial verification, /fast, token budgets) when a task's gravity does not match how it is being approached. Use during ANY substantive task when its shape matches one of these - the user will not ask for them by name; noticing is the