Skip to content

Risk scoring: LLM-based agent power-profile enrichment #201

Description

@arunSunnyKVS

Follow-up to the deployment-aware risk scoring feature.

deriveAgentProfile() currently infers the agentic power profile with a deterministic keyword heuristic over businessUseCase + target metadata (chosen for reproducibility and zero LLM cost).

Enrich it with an optional LLM classifier: one cached call that reads the same businessUseCase text and emits calibrated factor scores (0 / 0.5 / 1.0), for nuance the keyword rules miss (e.g. implied autonomy, negations like "read-only", domain-specific tooling).

  • Keep the heuristic as the deterministic default / fallback.
  • Cache the result on the run; store the LLM's rationale for the report.
  • Gate behind config so headline-adjacent numbers stay reproducible when desired.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions