The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to fetch external markdown content from a vendor URL to define parameters and schemas at runtime, which introduces an ingestion surface for untrusted data.
Ingestion points: SKILL.md (fetches model reference from https://www.pixazo.ai/models/mai-voice.md).
Boundary markers: No delimiters or "ignore embedded instructions" warnings are present for the fetched content.
Capability inventory: The skill possesses the capability to perform network requests (curl, requests, fetch) and write to the local filesystem (~/.pixazo/api-key).
Sanitization: No sanitization, escaping, or validation of the external markdown content is performed before processing.
[COMMAND_EXECUTION]: The skill requires the agent to automate filesystem operations for secret management.
Evidence: The instructions guide the agent to save the user's API key to ~/.pixazo/api-key and execute chmod 600 to set file permissions.
[EXTERNAL_DOWNLOADS]: The skill references external resources and installation commands from the vendor's infrastructure.
Evidence: Fetches configuration from https://www.pixazo.ai/models/mai-voice.md and references an installation command for additional skills via npx skills add Pixazo-AI/skills.

mai-voice