The Agent Skills Directory

[PROMPT_INJECTION]: The skill explores user-provided research repositories to extract findings and citations, which introduces a surface for indirect prompt injection if the files being processed contain malicious instructions.
Ingestion points: SKILL.md (Step 1: Explore the Repository workflow)
Boundary markers: Absent in the provided shell command examples.
Capability inventory: ls, find, grep, xargs, cat (all standard utilities).
Sanitization: Not explicitly implemented in the skill instructions; assumes the underlying agent platform handles shell argument sanitization.
[EXTERNAL_DOWNLOADS]: The skill references official conference assets and search tools from established providers.
Fetches official NeurIPS LaTeX styles from media.neurips.cc using a provided Makefile.
Recommends the installation of the Exa MCP search tool from mcp.exa.ai.
These sources are well-known and professional domains associated with the skill's primary purpose.
[COMMAND_EXECUTION]: The skill uses standard filesystem and search commands to facilitate the paper-writing workflow.
Employs find and grep to help the agent identify existing citations and research results in the user's codebase.
These operations are benign and expected for an agent skill tasked with analyzing research repositories.

ml-paper-writing