identity-drift-audit

To audit your Claude Code configuration, compare stored identity assumptions in CLAUDE.md and AGENTS.md against recent user behavior. The audit classifies claims as confirmed, stale, aspirational, wrong, or unverifiable. It then proposes concrete wording changes to update your personal configuration.

Identity drift occurs when AI assistant memory and configuration files retain outdated assumptions about user preferences and workflow habits. It happens when stored identity claims no longer match recent user behavior or workspace activity, requiring a claim-by-claim behavioral diff to detect mismatches.

To check if CLAUDE.md matches your workflow, run a behavioral diffing process that reads saved memory, skill settings, and hook configuration. This process compares those assumptions against recent messages and project activity, returning mismatches with proposed edits.

Yes, you can audit AGENTS.md and memory stores to find wrong assumptions. The audit extracts claims from these files and applies behavioral diffing against recent workspace activity, classifying each claim's accuracy and generating concrete proposed edits for correction.

The best way to update personal configuration is to generate an actionable report classifying each assumption as confirmed, stale, aspirational, wrong, or unverifiable. This report provides concrete proposed wording changes to directly update files like CLAUDE.md and AGENTS.md.

You should run an identity and preferences audit when your AI harness responses seem misaligned with your current goals. Run it to verify whether stored preferences, workflow habits, and memory still reflect your actual behavior rather than a stale snapshot.