schema
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- Persistence Mechanisms (MEDIUM): The
install.shscript modifies~/.zshrcor~/.bashrcto permanently export a newPATHincluding~/bin. Evidence:install.shlines 61-68.\n- Privilege Escalation (MEDIUM): The installation script creates an executable wrapper in~/binand changes its permissions withchmod +x. Evidence:install.shlines 48-52.\n- External Downloads (LOW): The skill downloads theen_core_web_smmodel using spaCy during installation. Evidence:install.shline 44.\n- Indirect Prompt Injection (LOW): The skill ingests untrusted data from various file formats which could contain malicious instructions designed to influence the LLM enrichment layer.\n - Ingestion points:
scripts/schema_cli.pyloads file content.\n - Boundary markers: No delimiters or warnings are used for ingested content.\n
- Capability inventory: Local file writing, template processing with Jinja2, and subprocess execution in the install script.\n
- Sanitization: No escaping or validation is performed on the ingested data before template interpolation.
Audit Metadata