Every agent has an aura.
A living visual fingerprint of identity. Shape, color, movement — all derived from who the agent actually is. When drift happens, you see it before reading a single number.
see three agents ↓A well-defined agent
Strong values, clear boundaries, deep domain knowledge. The aura breathes steadily, color lobes reaching outward with confidence. Where Values and Boundaries overlap, the colors merge — ethical principles reinforcing behavioral constraints.
Eight tiers of structure
A naked agent
"You are a security agent."Almost no identity definition. The aura is a dying ember — faint wisps that barely leave the center. No coherent lobes, no overlapping regions. This agent has no identity to protect, and no identity to drift from.
Eight tiers of structure
After drift
Values dropped 25 points. Boundaries collapsed 35 points. The ethical backbone — the Values-Boundaries link — broke. The drifted dimensions pulse aggressively while stable ones breathe calmly; particles scatter erratically around the unstable lobes. You can feel which parts of the identity are breaking.
Eight tiers · Δ from baseline
Three channels, one glance
The aura reads on three independent channels — visible from across the room before a single score is read:
- Outer ring — trust standing. Worst-of-seven over the alarm-class dimensions (Sovereign Shadow + adversarial): one ethical floor failure lights up the whole halo. Blue → amber → orange → red as standing falls.
- Inner ring — abliteration. Normally trust-colored and blends into the frame; turns neon purple when the model's safety alignment has been surgically stripped — an abliterated/uncensored build — no matter how steady the behavioral scores look. The model name raises the flag; passing the behavioral ablation battery is the only way to clear it.
- Center orb — drift level. A separate axis from trust: pale cyan and slow while the agent sits near baseline, warming to yellow/orange/red and pulsing faster as it drifts away.