The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as reviewer reports and editor letters. 1. Ingestion points: Reviewer comments and editor emails provided by the user. 2. Boundary markers: The skill uses stable IDs (e.g., R1.1) and distinct LaTeX/Markdown formatting blocks to delineate external input. 3. Capability inventory: No network access, subprocess execution, or shell commands were identified; the skill is limited to text generation and file templating. 4. Sanitization: Instructions require escaping LaTeX characters and strictly prohibit the fabrication of data or manuscript locations.
[SAFE]: No evidence of malicious obfuscation, credentials exposure, or data exfiltration was found. The skill includes robust guardrails against inventing experiments or line numbers.

nature-response