sag-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFE
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill manages ElevenLabs API keys via environment variables. It includes specific defensive instructions to prevent the agent from displaying, logging, or passing these keys as command-line arguments, which is a significant security improvement over standard implementations.
- [DATA_EXFILTRATION]: Explicit guardrails are included to prevent the piping of sensitive data or generated audio to external network commands like curl or wget.
- [EXTERNAL_DOWNLOADS]: The skill identifies the sag binary dependency and provides installation instructions via a specific Homebrew tap (steipete/tap/sag).
- [INDIRECT_PROMPT_INJECTION]: As the skill processes user-provided text for audio generation, it includes safeguards to require user confirmation for large or unbounded batches, protecting against potential cost-related attacks through injected instructions. Ingestion point: user message input in SKILL.md; Boundary markers: quoted shell arguments; Capability inventory: sag CLI execution; Sanitization: confirmation gates for bulk operations.
- [COMMAND_EXECUTION]: The skill utilizes the sag CLI for voice synthesis tasks, maintaining a local workflow for audio playback and file generation.
Audit Metadata