The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied source content (e.g., source.md or pasted text) and uses it to generate storyboards, character definitions, and image generation prompts. This represents an indirect prompt injection surface where malicious instructions embedded in the source content could attempt to influence the agent's behavior during the deep analysis or prompt generation phases.
Ingestion points: Source content is read from user-provided files or direct text input (Step 1.2).
Boundary markers: The skill templates (e.g., references/base-prompt.md) interpolate content without strict safety delimiters or specific instructions to ignore embedded commands.
Capability inventory: The skill possesses the capability to write files to the local filesystem (analysis, storyboard, prompts, and images), execute shell commands via bun for PDF merging, and invoke external image generation workflows.
Sanitization: There is no evidence of automated sanitization, filtering, or escaping of the input content before it is processed by the model. However, the workflow includes mandatory human review steps (Steps 4 and 6) where the user is prompted to approve the storyboard and prompts before final execution, which acts as a primary mitigation.

tuzi-comic