tzai-product

Pass

Audited by Gen Agent Trust Hub on Aug 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bash script found in dynamic paths using user-supplied parameters. This creates a surface for command injection.
  • Ingestion points: User-provided <kind> and <subject> arguments in SKILL.md.
  • Boundary markers: None. The script uses double quotes around variables, but does not sanitize for shell metacharacters like backticks or subshell expansions.
  • Capability inventory: Shell execution via bash in SKILL.md.
  • Sanitization: Absent; the skill expects the agent to provide clean inputs.
  • [EXTERNAL_DOWNLOADS]: The script suggests installing external code from the author's repository using npx if the local engine is not found. While targeting the same author's package (kedoupi/tzai-image-skill), this involves fetching and running code from an external registry.
  • [COMMAND_EXECUTION]: The skill dynamically searches for an execution engine across multiple hidden directories (.claude, .cursor, .grok, etc.) in the user's home folder and executes the first executable file it discovers without integrity verification.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 13, 2026, 02:21 AM
Security Audit — agent-trust-hub — tzai-product