The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content such as articles, web pages, and local Markdown files, which creates a surface for indirect prompt injection attacks. \n
Ingestion Points: The agent is instructed to read and analyze full article content from various sources, including potentially malicious external links and files, as defined in references/article-workflow.md. \n
Capability Inventory: The agent possesses capabilities to generate images, write character metadata and assets to the local file system, and modify existing local Markdown files. \n
Boundary Markers: The instructions do not define clear boundaries or delimiters to isolate untrusted input from the agent's core instructions. \n
Sanitization: No sanitization or safety checks are specified for the external content before it is processed by the agent. \n- [COMMAND_EXECUTION]: The skill requires the execution of a locally provided Python script to handle character registration and asset management. \n
Evidence: The skill instructions in SKILL.md and references/character-package.md direct the agent to run python3 scripts/character_registry.py with specific arguments to perform character lifecycle operations.

punk-ip-article-illustrations