The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content (text, files, or URLs) to generate comic storyboards and analysis, creating a surface for embedded instructions to potentially influence agent behavior.
Ingestion points: User-provided source content is saved to source-{slug}.md and processed during the content analysis phase described in references/workflow.md.
Boundary markers: The prompt instructions do not specify the use of clear delimiters or warnings to ignore instructions within the ingested source material during analysis.
Capability inventory: The skill leverages read_file, write_file, and terminal commands (curl, test) to manage assets and analyze content.
Sanitization: While the skill instructs the agent to strip secrets from source content, it does not include explicit measures to sanitize against prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The workflow involves automatically downloading image files from URLs returned by the image_generate platform tool using the curl command.
[COMMAND_EXECUTION]: The skill utilizes shell commands, specifically curl for downloading assets and test for verifying directory structures and file existence.

baoyu-comic