behavioral-segmentation

Pass

Audited by Gen Agent Trust Hub on Mar 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of reading and processing untrusted project files and behavioral data models. Evidence: 1. Ingestion points: Codebase files, customer event data, and profile structures accessed in Phase 1; 2. Boundary markers: The instructions lack explicit delimiters or 'ignore' warnings for the data being analyzed; 3. Capability inventory: The skill performs file read and write operations, including writing reports to 'docs/' and appending telemetry to '~/.claude/projects/'; 4. Sanitization: No sanitization or validation of the ingested content is defined.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 23, 2026, 10:56 AM
Security Audit — agent-trust-hub — behavioral-segmentation