Connect

Context Drift Check

ngreenwallbyngreenwall
•
2
•
Analytics & Monitoring

Audits session context and canary tokens to confirm Claude Code is actively following standing global instructions.

Context Drift Check is an observability and maintenance skill designed to detect silent rule loss in long Claude Code sessions caused by context compaction and memory decay. By utilizing depth-placed canary tokens and behavioral self-audits across the top, middle, and bottom of your global instructions (~/.claude/CLAUDE.md), it pinpoints exactly where context has dropped and issues a compact, 3-line diagnostic scorecard. Ideal for extended programming workflows, the skill can self-heal load issues via single-pass re-reads and highlights behavioral lapses so Claude can correct its own output in subsequent responses without tedious manual re-prompting.

Key Features

01Compaction tracking to distinguish natural conversation limits from unexpected dropped context
02Single-pass self-healing to reload instruction files without overwriting user data
03Self-auditing behavior checks that grade recent replies against defined core rules with verbatim proof
04Three-tier canary verification to detect loss across the top, middle, or bottom of instructions
05Concise scorecard output providing high-signal diagnostic status in under 40 lines
062 GitHub stars

Use Cases

01Diagnosing whether dropped rules stem from file loading issues or natural context limits
02Verifying that session-wide formatting, safety, or style constraints are strictly enforced
03Auditing instruction retention during long, multi-turn coding sessions prone to context compaction