nav-diagnose_skill

Overview

This skill detects drops in human–AI collaboration quality and triggers a short diagnostic and re-anchoring flow to restore alignment. It auto-invokes after repeated corrections, context mix-ups, loop stagnation, or explicit user signals like "something seems off." The goal is to catch misalignment early and re-establish a clear, shared understanding before productivity degrades.

How this skill works

The skill monitors recent exchanges (typically the last 10–15 messages) for quality signals: repeated corrections, hallucination reports, context confusion, frustration language, and loop stagnation. When thresholds are met, it calculates a severity level, displays a concise diagnostic summary, and prompts the user to confirm a reconstructed goal or correct assumptions. After confirmation it applies the re-anchored instructions and can optionally log diagnostics for adaptive preferences.

When to use it

• Auto-trigger after 2+ identical corrections on the same topic
• When user says phrases like "something seems off" or "you're not getting this"
• When hallucination or references to non-existent artifacts appear
• If loop mode shows 3+ identical state iterations (stagnation)
• When context usage is high and quality indicators start degrading

Best practices

• Do not trigger on a single correction or when user provides new requirements
• Keep diagnostics brief and non-blaming; focus on concrete observations
• Confirm specific points (goal, constraints, technical context) rather than general agreement
• Offer immediate corrective steps and preventive suggestions (e.g., compact context)
• Log recurrent issues to user profile to reduce repeat friction

Example use cases

• User corrects plural REST naming twice; skill re-anchors on naming convention
• AI mixes features from two projects; skill prompts technical re-alignment and suggests clearing context
• User shows frustration with repeated outputs; skill pauses, summarizes issues, and asks for correction
• Looped debugging step repeats; skill detects stagnation and offers next-step options
• Hallucinated file or function is reported; skill flags high severity and requests clarification