The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user prompts and reference image data, which represents a potential attack surface for indirect prompt injection.
Ingestion points: User text prompts and reference image URLs/base64 data (referenced in Step 3 and "Inputs").
Boundary markers: Absent; the skill does not use specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill performs network requests using curl, requests, and fetch, and writes to the local filesystem (~/.pixazo/api-key).
Sanitization: Absent; no sanitization of the input prompt or image content is performed before processing.
[EXTERNAL_DOWNLOADS]: The skill fetches technical reference documentation from a vendor-controlled external URL.
Evidence: The URL https://www.pixazo.ai/models/nano-banana.md is accessed to retrieve model parameter schemas and error codes.
[COMMAND_EXECUTION]: The documentation includes a command for managing related skills using the npx package runner.
Evidence: Mentions the command npx skills add Pixazo-AI/skills --skill '*' for adding further capabilities.

nano-banana