The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a set of shell scripts (image-provider.sh, bloom.sh, nano-banana.sh) to orchestrate image generation tasks, manage file system operations in the workspace, and interface with external APIs.
[DYNAMIC_EXECUTION]: The nano-banana.sh script implements a flexible fallback mechanism by executing a command defined in the NANO_BANANA_COMMAND environment variable, allowing the operator to provide custom adapters for alternative image generation workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from local strategy and brand files to generate image prompts, which creates a potential surface for indirect instruction injection if those source files are compromised.
Ingestion points: The skill reads data from workspace/pages/[page-name]/strategy.json, workspace/brand/profile.md, and workspace/pages/[page-name]/copy.md to inform its visual planning and prompting.
Boundary markers: There are no explicit delimiters or boundary markers used when interpolating content from the project files into the instructions or scripts.
Capability inventory: The skill possesses the capability to execute shell scripts, make network requests via curl to the Bloom API, and write files (visual plans, manifests, and image assets) to the project workspace.
Sanitization: Input data is handled through Python-based JSON parsing and environment variable resolution within the shell scripts to ensure that file paths and parameters are managed safely.

page-visuals