Filtering out unnecessary AI-generated work

★ 7.4 · vibe-coding

avoiding-manufactured-work is a Claude Code skill that detects and stops manufactured work — tasks, PRs, and artifacts created by an agent to appear productive rather than because a real consumer demanded them. It fires after a deliverable looks done and checks whether a concrete consumer exists: a user request, acceptance criterion, blocked downstream decision, or open reviewer thread. Each work product receives one of four dispositions: keep, shrink, defer, or delete — speculative and performative output is deleted, real but oversized scope is shrunk or deferred. The skill returns a structured four-field result: Disposition, Consumer, Reason, and Next action. It pairs with the `front-gate-before-pipeline` skill, which applies the same self-evaluation logic before work begins.