The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied academic text (LaTeX or prose) through an iterative critic loop. This presents an attack surface where malicious instructions embedded within a paper section could attempt to influence the agent's behavior.
Ingestion points: Phase 0 extracts paper text, equations, and citations from user input.
Boundary markers: No specific delimiters or safety instructions are defined to separate the input text from the skill's internal logic.
Capability inventory: The skill possesses the ability to execute shell commands (xelatex, grep) and perform file-system operations (reading references and writing constraints).
Sanitization: The skill lacks explicit sanitization or filtering logic for the input text prior to analysis.
[COMMAND_EXECUTION]: The skill uses local shell commands for text analysis and document validation. Specifically, Phase 6 and Phase 7 use grep to identify math symbols and xelatex/bibtex to verify the build integrity of the rewritten LaTeX sections. These operations are restricted to local processing and are consistent with the skill's primary purpose.
[DYNAMIC_EXECUTION]: The skill involves the generation and modification of LaTeX content followed by runtime compilation via xelatex. While this involves script execution, it is the core functionality intended for academic document validation and does not involve executing code from remote or untrusted external sources.

paper-section-rewrite