BitterPillEngineering

An AI prompt audit evaluates instruction sets to classify rules as CUT, RESOLVE, MERGE, EVALUATE, SHARPEN, MOVE, or KEEP. This prunes redundant scaffolding to improve output quality and provides estimated token savings.

Evaluating redundant rules in Claude AI instructions uses the Five Questions framework to assess necessity. It categorizes each rule to determine if it should be kept, cut, or sharpened, producing a report with estimated line and token savings.

Yes, you can run a QuickCheck workflow to perform a fast single-file audit. This provides a rapid keep, cut, or sharpen verdict for individual AI instruction files, compared to the full-system Audit workflow that reads all force-loaded files.

Synchronizing loadAtStartup and postCompactRestore.fullFiles in settings.json ensures the audit framework stays in sync with your configuration. This alignment is required so the full-system evaluation accurately reads all force-loaded instruction files.

The prompt audit framework relies on settings.json synchronization and requires force-loaded files to be properly configured. If loadAtStartup and postCompactRestore.fullFiles are not aligned, the full-system evaluation cannot accurately read or audit the instruction set.