The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is instructed to explore research repositories and read files such as READMEs and existing documentation (Ingestion points: SKILL.md identifies README.md and codebase exploration). These files are untrusted external data that may contain malicious instructions. The skill lacks explicit boundary markers or instructions to ignore embedded commands when processing these files. Given that the skill has capabilities for network requests (via Python requests) and filesystem operations, this creates a potential surface for indirect injection.
[EXTERNAL_DOWNLOADS]: The skill includes a build configuration in templates/neurips2025/Makefile that downloads LaTeX style files from the official NeurIPS conference domain (media.neurips.cc). This is an expected and legitimate operation involving a well-known research service.
[COMMAND_EXECUTION]: To support its workflow, the skill employs standard shell commands including find, grep, and pdflatex for repository exploration and document compilation. These actions are triggered by user-facing instructions and are appropriate for the skill's stated purpose.

ml-paper-writing