The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and summarize content from the user's research repository (code, results, and notes) and fetch paper abstracts from external APIs (Semantic Scholar, CrossRef, arXiv) to assist in drafting manuscripts. This ingestion of untrusted data creates a potential surface for indirect prompt injection if external content or local files contain malicious instructions.
Ingestion points: SKILL.md (instructions to explore repo via grep and find), references/citation-workflow.md (abstracts fetched from APIs).
Boundary markers: Absent; the skill lacks specific delimiters to isolate external text from drafting instructions.
Capability inventory: SKILL.md and templates/neurips2025/Makefile involve subprocess execution via shell commands (ls, find, grep, pdflatex, curl).
Sanitization: Absent.
[EXTERNAL_DOWNLOADS]: The templates/neurips2025/Makefile contains a command to download official conference style files directly from the NeurIPS conference domain (media.neurips.cc). This is standard behavior for academic templates to ensure they use the most up-to-date formatting requirements provided by the venue.

ml-paper-writing