The Agent Skills Directory

[OBFUSCATION]: Homoglyph obfuscation is present in scripts/generate_report_pdf.py and templates/report_latex.tex. The word 'amber' within the color name 'warnamber' is defined as 'warnambер', where the characters 'е' (U+0435) and 'р' (U+0440) are Cyrillic letters that appear identical to the Latin 'e' and 'r'. This technique is commonly used to bypass automated text-based security filters.
[COMMAND_EXECUTION]: The skill uses the Bash tool to locate files (find . -name "*.tex") and to execute a local Python script for report generation. This script (generate_report_pdf.py) invokes the pandoc command using subprocess.run to convert markdown into a PDF document.
[EXTERNAL_DOWNLOADS]: The skill requires the pymupdf library, as specified in requirements.txt. This package is used by the scripts/extract-pdf-annotations.py utility to read highlights and comments from PDF files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user-provided content from LaTeX source files and PDF annotations.
Ingestion points: SKILL.md (Step 0) instructs the agent to read all .tex files in a directory.
Boundary markers: Absent; there are no specific delimiters or instructions to the model to ignore embedded commands or instructions within the processed document text.
Capability inventory: The agent has access to Read, Bash, and Edit tools, creating a risk surface where malicious instructions in a processed paper could influence the agent to perform unintended file modifications or command executions.
Sanitization: Absent; the skill does not perform validation or sanitization of the content extracted from the input files before processing.

proofread