The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's instructions for generating schematics using python scripts/generate_schematic.py "your diagram description" facilitate a command injection vulnerability. Because the 'description' is derived from untrusted manuscript content, an attacker can embed shell metacharacters (e.g., ;, &, |) to execute arbitrary code when the agent runs the command.
[COMMAND_EXECUTION]: The skill explicitly requires and utilizes the Bash tool to run Python scripts and system commands, providing the necessary execution environment for the identified injection vulnerability.
[METADATA_POISONING]: The skill's metadata identifies the author as 'K-Dense Inc.', which contradicts the platform-reported author identity 'leonchaox'. This deceptive attribution can mislead users about the skill's origin and bypass reputation-based safety assessments.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted manuscripts.
Ingestion points: Manuscript text, grant proposals, and presentation files processed in SKILL.md.
Boundary markers: None present to distinguish data from instructions.
Capability inventory: Access to Bash, Read, Write, and Edit tools.
Sanitization: No sanitization of external content before interpolation into shell commands.
[EXTERNAL_DOWNLOADS]: The instructions reference external scripts (scripts/generate_schematic.py and skills/scientific-slides/scripts/pdf_to_images.py) that are not included in the provided file set, creating a dependency on unverifiable code that the agent is prompted to execute.
[PROMPT_INJECTION]: The skill uses high-pressure directives (e.g., '⚠️ CRITICAL', 'NEVER', 'MANDATORY') to override the agent's standard behavior, potentially forcing it into a specific workflow that utilizes vulnerable external scripts.

peer-review