The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The setup instructions include curl -fsSL https://bun.sh/install | bash. This pattern is a critical security risk as it executes a remote script with full shell privileges without any integrity verification. Since the source domain bun.sh is not within the defined trusted scope, this finding retains the highest severity.
[EXTERNAL_DOWNLOADS] (HIGH): The skill requires downloading and installing an external runtime (Bun) from a third-party source at runtime, which is not version-pinned or verified.
[CREDENTIALS_UNSAFE] (LOW): The documentation requests the user to set several environment variables for sensitive API keys (GOOGLE_API_KEY, REPLICATE_API_TOKEN, OPENAI_API_KEY). While these are required for operation, the presence of a remote execution vector increases the risk that these secrets could be harvested.
[INDIRECT_PROMPT_INJECTION] (MEDIUM): The skill ingests untrusted text via the --prompt flag to generate visual content.
Ingestion points: User-provided strings via the CLI --prompt flag in SKILL.md.
Boundary markers: None identified; prompts are interpolated directly into the tool execution.
Capability inventory: The skill performs network operations to external AI model APIs and writes files to the local filesystem (~/Downloads/).
Sanitization: No evidence of prompt sanitization or escaping is provided in the documentation.

Art