The Agent Skills Directory

Unverifiable Dependencies & Remote Code Execution (MEDIUM): The skill instructs the agent to run code from an external npm package that is not on the trusted sources list. Execution via npx downloads and runs the package immediately.
Evidence: Usage of npx -y superconductor-gemini-skills in all provided examples in SKILL.md.
Risk: If the npm package is compromised or maliciously crafted, it could execute arbitrary code on the host system.
Command Execution (LOW): The skill executes local shell scripts to interface with the Gemini API.
Evidence: bash ${CLAUDE_PLUGIN_ROOT}/scripts/gemini.sh.
Indirect Prompt Injection (LOW): The skill processes untrusted user-provided text for conversion to speech. While this primarily results in audio output, the surface for injection exists if the downstream agent uses the TTS results for further logic.
Ingestion points: TEXT TO SPEAK argument in shell and npx commands in SKILL.md.
Boundary markers: None present; the text is passed directly as a string.
Capability inventory: File system write (WAV file generation), network access (Gemini API).
Sanitization: None documented.
Credentials (INFO): The skill requires the GEMINI_API_KEY environment variable. This is a standard requirement for such services and is handled via environment variables rather than being hardcoded.

text-to-speech