Hook Audit Surfaces

Hook decisions that change model-visible context, submitted content, control flow, or permissions need a user-facing hook audit surface attached to the transcript artifact they affect: user turns, tool attempts/results, Agent tool rows, child agent initial prompts, or compaction summaries. Raw command-hook output and diagnostic runner logs stay out of the view by default, so hook authors must use structured decision fields when they want user-visible or model-visible effects. This keeps successful automation quiet while preserving auditability at the point where the hook changed what the model saw or what the workflow did.