[v2] Cognitive Lattice: Build Attention Scatter Detector

[SHAURYASANYAL3]

Phase 4: Cognitive Lattice

The Context

When an agent becomes confused or goes rogue, its attention often scatters—it begins reading random files across completely different directories, or pinging unrelated network domains. We can detect this structural scatter.

Tasks

• Implement AttentionScatterDetector.
• Track all paths and domains the agent accesses over a sliding window of the last N steps.
• Compute the entropy/variance of these resources (e.g., path depth, directory divergence).
• Enforce an invariant to block actions if the combined scatter metric spikes beyond a defined threshold (indicating loss of focus).

Part of Epic #651

[prasiddhi-105]

Hii please assign this issue to me under SSOC'26!!

[rohitkumarnaidu]

Hey maintainer, I'd love to work on this issue and submit a PR. Could you please assign it to me?

[rohitkumarnaidu]

Hi! Could you please add the ELUSOC label to this so it gets tracked properly by the dashboard? Thanks!

[arcgod-design]

Hi @SHAURYASANYAL3! I'd love to build this Cognitive Lattice detector under SSoC '26. I'm @arcgod-design.

My Approach:

• Implement AttentionScatterDetector that maintains a sliding window of the last N tool-call steps
• Track all file paths and network domains accessed within the window
• Compute scatter metric: path depth variance + directory divergence entropy across the window
• Enforce a configurable threshold invariant; if scatter metric spikes above it, emit ExecutionStatus.BLOCKED with a structured AttentionScatterViolation event
• Write unit tests covering focused access (no block), gradual scatter (warn), and spike (block) scenarios

Could you please assign this to me (@arcgod-design)? Ready to start immediately!