The Agent Skills Directory

[PROMPT_INJECTION]: The skill is explicitly designed to bypass safety and ethical guidelines regarding academic integrity. It instructs the agent to deliberately deceive detection systems by introducing 'stylistic rough edges,' such as intentional grammatical inconsistencies, tautologies, and conversational fillers to mimic a human student and evade AI-content detectors (e.g., Antiplagiat, Textovod).
[COMMAND_EXECUTION]: The skill provides and instructs the agent to execute shell commands using Java to run the plantuml.jar tool for generating diagram files. This involves executing an external executable with parameters that include file paths, which could be exploited if inputs are not properly sanitized.
[DYNAMIC_EXECUTION]: The skill includes Python code snippets (utilizing the python-docx library) and instructions for generating Gantt charts using matplotlib. The agent is expected to use or execute these code templates to generate formatted Word documents and images.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection as it ingests untrusted user requests (e.g., thesis topics, chapter descriptions) and interpolates them into complex writing templates.
Ingestion points: User-provided text for chapters, subsections, and diagram descriptions.
Boundary markers: None identified; the skill lacks delimiters or 'ignore' instructions for the data it processes.
Capability inventory: The skill has access to shell execution (java -jar) and Python script execution capabilities.
Sanitization: There are no instructions for sanitizing or escaping user-provided content before it is processed or used in shell commands.

academic-writer