The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run within scripts/format_guard.py to invoke LaTeX compilers (specifically xelatex) and bibtex. These calls are used to verify project build integrity after modifications and are executed as controlled lists of arguments restricted to the project root, rather than through a shell.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local project data which creates an injection surface.
Ingestion points: scripts/resource_scanner.py reads filenames and content from code (.py, .c, etc.), references (.bib), and figures. scripts/semantic_analyzer.py reads existing .tex files.
Boundary markers: Prompts in config.yaml utilize structural headings but do not include explicit 'ignore embedded instructions' warnings for the interpolated resource data.
Capability inventory: scripts/format_guard.py performs file writes for project modifications and executes compiler subprocesses. scripts/skill_controller.py manages file system operations for backups and logs.
Sanitization: scripts/security_manager.py implements proactive sanitization, filtering out dangerous LaTeX commands (e.g., \geometry, \setlength, \usepackage) and keywords from generated content before application.
[EXTERNAL_DOWNLOADS]: The skill documentation in SKILL.md instructs the agent to use the gh (GitHub) CLI to upload bug reports to the vendor's repository at huangwb8/bensz-bugs. This is a standard vendor-provided maintenance channel and is governed by constraints prohibiting the upload of sensitive credentials or identity information.
[SAFE]: The skill implements several defensive best practices, including isolated task directories (.bensz-api/ and .complete_example/), SHA256 integrity verification for system configuration files, and a custom template renderer in scripts/template_renderer.py that avoids unsafe evaluation methods.

complete-example