The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts external input through frame specifications and direct prompt strings, which are then used as instructions for remote image generation models.
Ingestion points: Input is accepted via the --spec and --prompt arguments in scripts/lib/generate-image.mjs.
Boundary markers: The skill does not implement delimiters or specific instructions to isolate user-provided text from the generation task.
Capability inventory: The skill can write files to the local filesystem (images, metadata, and logs) and perform network operations to the OpenRouter API.
Sanitization: No sanitization or validation is applied to the input prompt data before it is sent to the model provider.
[COMMAND_EXECUTION]: The skill utilizes shell script wrappers to execute Node.js implementation files that interact with the system. These scripts perform operations including reading configuration, writing image files, and creating directories to store usage logs.
[DYNAMIC_EXECUTION]: The skill performs dynamic module loading using paths computed at runtime to integrate with shared vendor libraries.
Evidence: In scripts/lib/resolve.mjs and scripts/lib/ledger-bridge.mjs, the skill uses dynamic import() calls to load functionality from sibling directories (e.g., studio-core, studio-universe, studio-ledger) based on environment variables or relative path calculations.

studio-imagegen