The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from various sources, including user-provided text, external URLs, and local file paths, to generate comic storyboards and image prompts. There is an inherent risk that malicious instructions embedded in this source material could attempt to hijack the agent's behavior during the analysis or generation phases.
Ingestion points: User-provided content, URLs, and file paths used for comic generation (SKILL.md, Step 1.1).
Boundary markers: The instructions do not mandate the use of delimiters or specific "ignore" instructions when the agent processes the ingested content.
Capability inventory: The skill utilizes write_file, read_file, image_generate, and the system terminal (for curl operations).
Sanitization: While the skill includes a "Strip secrets" directive to scan for credentials in source content, it lacks formal sanitization or validation of the data before it is interpolated into prompts.
[COMMAND_EXECUTION]: The skill uses the system terminal to execute curl commands for downloading image assets returned by the generation tool. The commands are constructed using dynamic variables such as kebab-case slugs derived from the input topic. If the topic name is manipulated to include shell metacharacters, it could potentially lead to command injection.
[EXTERNAL_DOWNLOADS]: The workflow involves downloading image files from remote URLs (returned by the image_generate tool) to the local file system. This is a functional requirement for persisting the comic pages, though it represents a standard external network dependency.

baoyu-comic