The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/latex_checker.py invokes clean_latex.py using subprocess.run() when the --fix flag is supplied. This execution is implemented safely by passing parameters as a structured list without enabling shell parsing, which prevents shell command injection through malicious file paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data by reading external LaTeX documents.
Ingestion points: Files are ingested via load_tex() in scripts/latex_checker.py and direct file reads in scripts/clean_latex.py.
Boundary markers: No delimiters or isolation techniques are applied to text snippets extracted from the document before they are printed.
Capability inventory: The skill performs local file read/write operations and executes a local script via a safe subprocess call. No network access or dynamic language evaluation primitives are present.
Sanitization: Text fragments from the input file (such as TODO comments or author fields) are displayed directly to stdout without sanitization, creating an indirect prompt injection surface if an untrusted document contains instructions intended to manipulate the agent's behavior.

latex-formatting