The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, including academic manuscripts, Chinese-to-English drafts, LaTeX source code, and PDF files.\n
Ingestion points: User-provided prose text and PDF files (processed via libraries like pymupdf and pypdf) are the primary inputs for the polishing and layout diagnosis workflows.\n
Boundary markers: The instructions do not specify the use of boundary markers (e.g., XML tags or clear delimiters) to separate user content from system instructions during the analysis.\n
Capability inventory: The skill provides logic for restructuring text and suggests rendering/compiling operations for LaTeX documents and PDFs.\n
Sanitization: No explicit sanitization or instruction-filtering for the ingested academic text is mentioned to prevent embedded commands from being followed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill manifest and journal-specific fragments reference shared resources located outside the skill's primary folder.\n
Evidence: The manifest.yaml and static/fragments/journal/nat-comms.md reference files in a ../_shared/ directory structure (e.g., ../_shared/core/reader-workflow.md).\n
Context: These represent local infrastructure dependencies within the skill's author environment rather than remote downloads from untrusted third-party sources.

nature-polishing