The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script scripts/cip/render-html.py injects user-provided data, such as brand names and industry descriptions, directly into an HTML template without any sanitization or escaping. This creates a cross-site scripting (XSS) vulnerability surface where malicious input could execute scripts if the generated HTML is opened in a browser.
[DYNAMIC_EXECUTION]: The skill dynamically generates executable content in the form of HTML presentations and SVG icons. In scripts/icon/generate.py, SVG code is extracted from AI responses and saved to files; since SVGs can contain embedded scripts, this poses a risk if the AI model is influenced to include malicious payloads.
[PROMPT_INJECTION]: SKILL.md contains an instruction: 'IMPORTANT: When scripts fail, try to fix them directly.' This encourages the agent to modify the skill's own codebase at runtime, which is a form of self-modification that could be exploited if an attacker manages to trigger a script failure with controlled input.
[COMMAND_EXECUTION]: The skill relies extensively on the execution of Python scripts to perform its core functions, such as scripts/logo/generate.py and scripts/cip/generate.py, which process user-provided arguments into shell commands.
[EXTERNAL_DOWNLOADS]: The generate.py scripts for logo and CIP generation include logic to download image files from external URLs provided by API services (Gemini, Atlas Cloud, and MuAPI). While the implementation includes hostname and IP validation to mitigate SSRF risks, it remains an external network dependency.

design