The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages a PIXAZO_API_KEY by reading from environment variables or a local file. It instructs the agent to save keys to ~/.pixazo/api-key using chmod 600, which is a standard security practice for credential management.\n- [COMMAND_EXECUTION]: The skill uses system commands (curl, chmod) to interact with APIs and manage file permissions. These operations are scoped to the skill's intended functionality.\n- [EXTERNAL_DOWNLOADS]: The skill fetches documentation and polling configurations from the vendor's official domains (www.pixazo.ai) and uses image assets from a Cloudflare R2 bucket (r2.dev).\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user prompts to generate video content. \n
Ingestion points: User-provided prompt and image URL in the API request body in SKILL.md.\n
Boundary markers: Absent.\n
Capability inventory: Uses curl, requests, and fetch for network operations, and chmod for file system management.\n
Sanitization: Absent.\n
Conclusion: While the ingestion of untrusted data represents an attack surface, the skill lacks dangerous autonomous capabilities or privilege escalation vectors that could be exploited.

sora