The Agent Skills Directory

[SAFE]: The skill provides instructions for managing API credentials and interacting with vendor-controlled endpoints. It emphasizes security best practices, such as using environment variables instead of hardcoded secrets and requiring user acknowledgment before data transmission.
[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for handling user-provided media and text inputs for AI model inference. This represents an indirect prompt injection surface where external content is processed by the agent.
Ingestion points: User-provided local files and prompts are processed in references/pruna-api.md for transmission to external APIs.
Boundary markers: The skill relies on the agent's default prompt handling and does not specify additional delimiters for user input.
Capability inventory: The skill utilizes curl for file system operations (reading files for upload, writing output) and network requests to api.pruna.ai and api.replicate.com as documented in references/pruna-api.md and references/replicate-api.md.
Sanitization: No explicit sanitization of the content within user-provided media or prompts is performed locally, as filtering is typically handled by the target model providers.

pruna-api