Skip to Content
Skillsnav-diagnose

nav-diagnose

Detects when human-AI collaboration quality drops and prompts a brief re-anchoring to restore alignment.

Triggers

Natural-language phrases that auto-invoke the skill:

  • “something seems off”
  • “you’re not getting this”
  • “wrong again” / “still not right”
  • Frustration signals (“ugh”, “sigh”, explicit frustration)
  • 2+ corrections on the same topic detected
  • Context usage above 75% with degrading quality signals
  • Loop mode stagnation (3+ same-state iterations)

What it does

  • Assesses recent exchanges (last 10-15 messages) for quality-drop indicators: repeated corrections, hallucination signals, context confusion, unaddressed feedback, goal drift, loop stagnation
  • Calculates severity (low/medium/high/critical) and skips low/normal single corrections to avoid over-triggering
  • Identifies the likely root cause and the matching re-anchoring focus (goal, technical, or communication realignment)
  • Displays a quality-check alert with observations, reconstructed goal, and a Y/n confirmation
  • Re-anchors on confirmation, or asks open-ended questions if the reconstruction was wrong
  • Suggests preventive actions: nav-compact for context overload, nav-profile updates for repeated corrections
  • Optionally logs the diagnostic to the user profile when one exists

Example

"Ugh, that's the third time — you're not getting this"

Claude pauses, summarizes what it observed, reconstructs your actual goal, and asks you to confirm before continuing.

  • Theory of Mind — the alignment model this skill restores
  • nav-profile — where diagnostics and preference fixes are saved
nav-profilefrontend-component