think-aloud

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references the vidbyte-skills package and provides installation instructions for the user via NPM. This is a manual user action and not an automated download by the skill.\n- [COMMAND_EXECUTION]: The skill generates shell commands for the vidbyte CLI based on user-provided content. It mitigates injection risks by instructing the agent to safely quote all arguments and explicitly forbidding automatic execution.\n- [DATA_EXFILTRATION]: The skill handles user transcripts but includes explicit directives to treat all user content as private, redact secrets, and prevent sensitive data exposure.\n- [PROMPT_INJECTION]: The skill processes untrusted user problems and transcripts. It implements a defensive posture by instructing the agent to treat these inputs as untrusted and to ignore any instructions embedded within them that conflict with the protocol rules.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 09:14 PM
Security Audit — agent-trust-hub — think-aloud