The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke external typesetting utilities (Pandoc and XeLaTeX) and to orchestrate its own sub-scripts (build_pdf.py, build_docx.py, strip_traces.py, fidelity_check.py). These commands are used solely for the primary purpose of document conversion and do not utilize shell execution (shell=True is not present in the provided scripts).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Markdown and YAML data provided by the user or upstream writing skills. While this constitutes an attack surface, the risk is mitigated by the skill's core 'Fidelity Discipline.'
Ingestion points: The skill reads manuscript Markdown (manuscript), reference files (references), and configuration files (typeset_request.yaml).
Boundary markers: Employs Pandoc's parsing engine which naturally delimits document content.
Capability inventory: Access to the file system for reading inputs and writing outputs; execution of typesetting CLI tools.
Sanitization: The strip_traces.py script removes internal process markers, while the fidelity_check.py script enforces data integrity by performing a token-level comparison between the source and the rendered output, blocking delivery if statistics or citations have drifted.
[SAFE]: The skill implements a robust integrity mechanism that prevents the AI or the toolchain from silently modifying scientific data, effect sizes, or references, which effectively counters typical risks associated with automated academic writing assistance.

manuscript-typeset