The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The error recovery section in resources/execution-protocol.md instructs the agent to suggest a piped remote execution command (curl -LsSf https://astral.sh/uv/install.sh | sh) to the user for tool installation. While this is an official installation method for the uv tool, the pattern itself involves executing remote scripts directly in the shell.
[EXTERNAL_DOWNLOADS]: The skill relies on uvx to dynamically download and execute packages from the Python Package Index (PyPI), including opendataloader-pdf, opendataloader-pdf-hybrid, and mdformat. These are external dependencies fetched at runtime.
[COMMAND_EXECUTION]: The skill performs multiple shell operations to fulfill its purpose, including file metadata inspection (wc, ls, pdfinfo) and the execution of conversion utilities via uvx.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it processes untrusted external data.
Ingestion points: Untrusted PDF content is read from user-provided paths (input_path) as described in SKILL.md and resources/execution-protocol.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are used during the extraction or formatting phases.
Capability inventory: The skill has the ability to execute subprocesses via uvx and perform file system read/write operations.
Sanitization: There is no evidence of sanitization, filtering, or escaping of the content extracted from the PDF before it is presented to the agent or saved as Markdown.

oma-pdf