context-audit

Audit conversation context by inventorying loaded CLAUDE.md files, rules, skill definitions, file reads, tool results, and conversation turns to identify the largest contributors to context usage. The audit generates a structured report detailing top space consumers and detected waste types.

Context window waste detection identifies stale reads, oversized outputs, duplicated context, and unused data occupying valuable space. This process flags unnecessary items consuming context bandwidth, enabling targeted compaction or load shedding to optimize long-running sessions.

Reduce context usage in long-running sessions by applying actionable recommendations from the audit report, which proposes compacting existing context or shedding load. The Skill requires explicit user confirmation before applying any context reduction changes.

Identify specific file reads and tool outputs consuming context space by running an audit on the active session. The resulting structured report details the largest contributors to context usage, highlighting specific items for potential removal or compaction.

Start a fresh conversational AI session when the audit report recommends it, specifically when context reduction through compaction is insufficient to reclaim necessary bandwidth. This action is suggested alongside compact or focused context reduction options during long-running sessions.

The context audit does not automatically delete unused data from the session. It outputs a structured report with recommended actions for compaction or shedding load, requiring explicit user confirmation before applying any changes to the context window.