The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text to generate image prompts, which creates a potential surface for indirect prompt injection where malicious instructions in the data could affect the agent's behavior.
Ingestion points: User-supplied subjects, product details, and literal text strings for images, as described in the "Generation flow" section of SKILL.md and "Domain configurations" in references/domain-configurations.md.
Boundary markers: The skill lacks explicit delimiters or "ignore instructions" wrappers for the user-provided content within the API request prompts.
Capability inventory: The skill executes curl requests to the external endpoint at api.pruna.ai (SKILL.md).
Sanitization: No evidence of sanitization or validation of external content is present before its interpolation into the prompt.
[EXTERNAL_DOWNLOADS]: The skill manages dependencies by downloading additional components from the author's official repository.
Evidence: SKILL.md contains commands such as npx skills add PrunaAI/pruna-skills@pruna-api -y to install required prerequisites.
Context: These references are targeted at the vendor's own pruna-skills repository and represent legitimate functional extensions.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands for API interaction and skill management.
Evidence: SKILL.md includes curl templates for making predictions at https://api.pruna.ai/v1/predictions and npx commands for adding new skills.

p-image-ideogram