skills-agents-sweeplisted

# Skills and Agents Sweep Skill Implements `spec/claude/skills-agents-sweep/` — the spec defines triggers, scope, phases, and lifecycle. This skill binds those rules to the on-disk procedure. The sweep supplements per-artefact reviews with cross-cutting analysis that no single per-artefact review can surface: boundary conflicts between two skills only emerge when both are reviewed together, spec-induced gaps (phantom skills) are invisible to a review of existing artefacts, and operations-vocabulary drift accumulates across the inventory without any individual finding calling it out. ## Why this is a skill, not an agent - **Mid-flow interactivity** — scope confirmation (full inventory vs. narrowed subset by phase or tag), wave-decision sign-offs (implement vs. defer vs. retire), and consolidated-report approval all require user input at multiple checkpoints. An agent's fire-and-forget contract loses that deliberation loop. - **Persistent on-disk output is the contract** — the consolidated report under `.audits/skills-agents-sweep/` must survive past the current turn and be referenced by every downstream implementation PR as its evidence source. Skills own persistent artefacts; agents return ephemeral reports. - **Orchestration role** — this skill dispatches `skill-review` and `agent-review` as sub-procedures for phase 1 and chains to `pull-request-create` when waves are committed. The skill-orchestrates pattern defaults the orchestrator to skill form per `spec/claude/skill