user-personalization-learner

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructions require the agent to record highly sensitive Personal Identifiable Information (PII), specifically NPWP/NIK (16-digit national IDs) and business classification details. While necessary for tax assistance, instructions to store this in an Obsidian vault or JSON files for other tools create a localized data exposure surface.
  • [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by consuming arbitrary user data via the /learn command and storing it in a "long-term memory" used by other agents.
  • Ingestion points: User-supplied strings processed via the /learn slash command as described in SKILL.md.
  • Boundary markers: Absent. The instructions do not define delimiters or specific warnings to ignore instructions embedded within the user-provided preference data.
  • Capability inventory: The ingested data is explicitly shared with other capabilities such as tax-report-generator and obsidian-tax-vault-manager (file system sync).
  • Sanitization: Absent. There is no mention of validation, filtering, or sanitization of the data before it is recorded into the system's memory.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 09:15 AM
Security Audit — agent-trust-hub — user-personalization-learner