The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using uvx and uv run. These commands are used for converting PDF files to markdown/images and for searching citation databases.
[EXTERNAL_DOWNLOADS]: The skill invokes uvx opencite, which downloads and executes an unversioned package from an external registry at runtime. The opencite tool is not from a verified trusted organization.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves processing untrusted academic manuscripts (PDFs) and using content derived from them to perform external tool calls. 1. Ingestion points: Untrusted manuscript PDF files and user-supplied keywords derived from the text (references/review-procedure.md). 2. Boundary markers: None. The instructions do not specify how to delimit untrusted manuscript content from the agent's instructions, nor do they include warnings to ignore embedded instructions within the papers. 3. Capability inventory: The agent has access to shell execution (uvx, uv run), file system writes (saving converted images and markdown), and network capabilities (via citation searches). 4. Sanitization: There is no mention of sanitizing or escaping manuscript text before it is used to generate search queries or processed by conversion tools.
[DYNAMIC_EXECUTION]: The skill includes an embedded Python script in references/review-procedure.md that is executed at runtime via uv run to handle PDF-to-image conversion. While the script is a utility for image processing, it represents a pattern of executing dynamically generated code.

paper-review