The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install software from external sources using 'pip install opendataloader-pdf' and 'npm install @opendataloader/pdf'. It also references external GitHub repositories hosted by 'opendataloader-project' which are not among the pre-approved trusted sources.- [COMMAND_EXECUTION]: The skill provides CLI commands to execute the parser and indicates that the tool spawns JVM subprocesses during operation. It also involves starting a background server on port 5002 for hybrid processing modes.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted PDF documents and convert them into Markdown, JSON, or HTML for use in RAG and LLM pipelines.
Ingestion points: Processes external PDF files via 'input.pdf' or 'input_path'.
Boundary markers: No explicit instruction delimiters or 'ignore embedded instructions' warnings are provided for the extracted content.
Capability inventory: The skill executes subprocesses (JVM), performs file writes to an output directory, and runs a network server.
Sanitization: No sanitization or filtering of the extracted text content is mentioned before it is passed to the AI agent.
[DYNAMIC_EXECUTION]: The documentation mentions that running 'npm run sync' will regenerate Python and Node.js bindings based on Java option changes, indicating a pattern of local code generation.

opendataloader-pdf