The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted user-provided text (articles, pasted content) to generate outlines and prompts for an external image generation tool. Malicious instructions embedded in the input content could potentially influence the agent's analysis or the resulting image generation.\n
Ingestion points: User-provided text and files processed in SKILL.md and references/workflows/analysis-framework.md.\n
Boundary markers: Prompt templates in references/workflows/prompt-assembly.md use section markers like {CONTENT_SECTION}, but do not provide explicit instructions to the agent to ignore any embedded directives within the user data.\n
Capability inventory: The skill can read/write local files and call an image generation tool (nano banana pro).\n
Sanitization: No sanitization or escaping of the user-provided content is performed before interpolation.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell and PowerShell commands to check for the existence of its local configuration file (EXTEND.md) and determine its location. These commands are limited to path existence checks and do not involve remote access or sensitive data operations.\n
Evidence: File check logic in SKILL.md using test -f and Test-Path.

baoyu-image-cards