The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of local Python scripts (scripts/generate_schematic.py, scripts/generate_image.py) using arguments derived from dynamically generated descriptions of figures and diagrams.
[COMMAND_EXECUTION]: The skill instructs the agent to compile generated LaTeX source code using the xelatex command, which constitutes the execution of locally generated code.
[PROMPT_INJECTION]: Imperative language and symbols such as "⚠️ MANDATORY" and "⚠️ CRITICAL" are used to force the use of specific figure-generation tools, which can serve to override agent autonomy and bypass contextual decision-making.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes external literature data retrieved via the research-lookup skill into the final manuscript output.
Ingestion points: Research data and citations retrieved via the research-lookup skill referenced in the workflow (SKILL.md).
Boundary markers: The skill does not implement delimiters or instructions for the agent to ignore potentially malicious directives embedded in the external research data.
Capability inventory: Shell command execution (Python scripts) and compilation of LaTeX source code.
Sanitization: No sanitization, validation, or filtering procedures for ingested external content are described.
[PROMPT_INJECTION]: Metadata poisoning is identified through a discrepancy between the author declared in the YAML frontmatter ("K-Dense Inc.") and the provided author context ("sickn33"), along with an unusual and potentially deceptive "risk: critical" field.

scientific-writing