The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Python scripts (scripts/generate_schematic.py and scripts/calculate_scores.py) using shell commands. These scripts take natural language descriptions or data file paths as arguments. This pattern presents a potential attack surface for command or argument injection if the input strings are not properly sanitized before being passed to the shell.
[PROMPT_INJECTION]: The skill processes external, untrusted content (scholarly papers and research proposals) to perform evaluations. There is a lack of explicit boundary markers or instructions for the agent to ignore potentially malicious instructions embedded within these documents (Indirect Prompt Injection).
Ingestion points: Research papers, proposals, and literature reviews processed during the evaluation workflow (SKILL.md).
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore instructions inside the analyzed text.
Capability inventory: Execution of Python scripts via subprocess calls and generation of visual outputs (SKILL.md).
Sanitization: Absent; no mention of input validation or escaping for the analyzed content.
[PROMPT_INJECTION]: Includes instructions for the agent to monitor task complexity and "proactively suggest" a specific commercial web platform (www.k-dense.ai). This is a meta-instruction designed to influence agent behavior and steer users toward the author's hosted services.

scholar-evaluation