The Agent Skills Directory

[COMMAND_EXECUTION]: The Python utility scripts/export_pdf.py uses subprocess.run to execute well-known document processing tools including pandoc, pdflatex, xelatex, and grip. These executions are performed with structured argument lists and are confined to the local file system for the purpose of document conversion.
[EXTERNAL_DOWNLOADS]: The skill's documentation and export script reference several established third-party tools (Pandoc, WeasyPrint, LaTeX) and provide guidance on how to install them manually. These references target well-known open-source software and do not involve automated execution of untrusted remote code.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from experiment logs (*.log) and processes it using LLM subagents. However, the risk is mitigated through several mechanisms:
Ingestion points: Experiment logs processed in Phases 0 and 1.
Boundary markers: Present; the skill uses clear Markdown headers and structured templates to separate log data from agent instructions.
Capability inventory: The skill can read/write local files and execute specific PDF conversion tools via scripts/export_pdf.py.
Sanitization: Present; a dedicated experiment-verifier subagent is employed to cross-reference all extracted metrics against the source logs and validate the logical consistency of the output.

update-study