The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional Markdown and YAML files without any executable code, shell scripts, or binary dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted reviewer comments and author notes, creating an indirect injection surface.
Ingestion points: Reviewer comments and editor letters provided by the user in the prompt as defined in SKILL.md.
Boundary markers: The workflow in static/core/workflow.md requires segmenting comments into stable IDs (e.g., R1.1) to isolate and track individual instructions.
Capability inventory: The skill does not possess execution capabilities such as subprocess calls, file-system writes, or network operations.
Sanitization: The references/qa-checklist.md and static/core/stance.md enforce strict verification protocols, prohibiting the fabrication of scientific content and mandating placeholders for missing author input.
[COMMAND_EXECUTION]: No shell commands or system calls were identified. The skill does not utilize dynamic context injection (!command syntax).
[DATA_EXFILTRATION]: No network-enabled tools or scripts are present. References to external URLs in references/source-basis.md are limited to official academic publishing domains (nature.com, springernature.com) and serve as documentation sources.

nature-response