The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/template_downloader.py script downloads document templates and style guides from well-known academic and medical domains, including nature.com, sciencemag.org, cell.com, ieee.org, and jamanetwork.com. These are established sources consistent with the skill's stated purpose.
[COMMAND_EXECUTION]: In scripts/format_converter.py, the skill executes pandoc --version via a hardcoded subprocess call. This is a safe operation used to verify that the required system tool is installed and available.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document content provided by the user to perform format adjustments.
Ingestion points: Local files (.docx, .md, .tex) specified via the --input flag in scripts/init_run.py.
Boundary markers: None; document text is parsed directly for formatting markers.
Capability inventory: The skill can write files to the local system and make network requests to fetch journal templates.
Sanitization: Standard document parsing libraries (python-docx, markdown) are used, but the skill does not explicitly sanitize text for embedded instructions.
[METADATA_POISONING]: The skill includes a pre-packaged audit file article-format-adjustment_audit_result_v1.json which contains safety assertions and a "PASS" verdict. Per security rules, these self-referential claims are treated as evaluated data rather than authoritative conclusions about the skill's actual security posture.

article-format-adjustment