The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages service credentials stored at ~/.config/paprika3d/credentials.json. The instructions contain explicit safeguards, directing the agent to never read, print, or display the contents of this file to prevent credential leakage.
[COMMAND_EXECUTION]: To prepare projects for publishing, the agent is instructed to execute npm install and vite build within the project directory. This involves running code defined in the project's configuration and dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes files from the user's project directory to facilitate the build and publishing process, creating a surface for potential injection via project content. Ingestion points: Project files and folders located at the cwd (Current Working Directory) specified during the publish tool call (SKILL.md). Boundary markers: The instructions do not define specific delimiters or warnings to separate untrusted project content from the agent's internal logic. Capability inventory: The skill utilizes shell command execution (npm install, vite build) and network communication to the paprika3d.app domain (SKILL.md). Sanitization: Sanitization is limited to the agent reviewing warnings returned by the publish tool regarding technical link or build errors.

paprika-publish