AI Persona Engine
PassAudited by ClawScan on May 10, 2026.
Overview
This instruction-only skill appears coherent and benign, but its optional learning-loop design may process call transcripts and store audit data or prompt patches for future persona behavior.
Safe to treat as a prompt-design aid. Before implementing the learning loop, confirm where transcripts and audits are stored, who can access them, how long they are retained, and whether generated prompt patches require human approval before affecting future personas.
Findings (1)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If implemented, call-derived quotes or learned prompt patches may persist and shape later persona responses.
The design processes live-call transcript data, stores audit records, and reuses prompt patches at runtime, creating persistent context that can influence future persona behavior.
Live Call → fetch_transcript → analyze_call → audit_persona → persona_audits (Supabase) ... prompt_patches table → persona_engine reads at runtime
Use only consented transcripts, minimize stored quotes, set access controls and retention limits, and review or validate prompt patches before enabling them at runtime.