The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run() to execute local binaries for LaTeX compilation (xelatex, bibtex), document conversion (soffice, osascript, qlmanage), and helper scripts.
Evidence in scripts/enhanced_optimize.py, scripts/generate_baseline.py, and scripts/core/validators/compilation_validator.py.
The skill implements a WorkspaceManager that resolves project paths and ensures they are located within the designated projects/ directory using Path.resolve() and relative_to(), which mitigates path traversal risks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it reads text content from external PDF and LaTeX files and interpolates this data directly into prompts for the AI reasoner.
Ingestion points: scripts/analyze_pdf.py (extracts text from PDF) and scripts/extract_headings.py (extracts text from Word/LaTeX).
Boundary markers: None identified; the text is placed into templates in scripts/core/decision_reasoner.py without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: High. The skill can modify configuration files and execute shell commands for compilation.
Sanitization: Not present for the ingested text.
While this is an attack surface, it is inherent to the tool's primary purpose of document analysis and is considered a low risk in this context.

make-latex-model