The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates external data (user descriptions and document content) into prompts for the underlying image generation model, creating a potential surface for indirect prompt injection.
Ingestion points: Untrusted data enters via the --desc and --context arguments in SKILL.md and the corresponding parameters in scripts/generate.mjs.
Boundary markers: The skill uses Markdown templates (scripts/prompts/user.md) with distinct headers (e.g., 'Description', 'Document Content') to separate instructions from user-provided data.
Capability inventory: The skill can write files to the local file system and perform network requests to download generated images.
Sanitization: No specific text sanitization or escaping is applied to the input variables before interpolation into the prompt templates.
[EXTERNAL_DOWNLOADS]: The skill uses curl to download generated images from remote URLs returned by the AFS CLI backend. These downloads are integral to the skill's purpose and are fetched from service endpoints managed by the vendor.
[COMMAND_EXECUTION]: The skill executes several shell commands for environment setup and core functionality.
It performs backend detection by checking environment variables and executing which afs and ls commands.
It invokes node to run its internal generation script and afs exec to interface with the vendor's image generation actions.

doc-smith-images