The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/mineru_parse.py uses the os.system() function to execute shell commands for environment setup and model management.
Evidence: os.system("pip install uv && uv pip install -U 'mineru[all]'") and os.system("mineru-models-download") in scripts/mineru_parse.py.
[EXTERNAL_DOWNLOADS]: The skill triggers the download of external Python packages and large machine learning model weights at runtime.
Evidence: Setup instructions in SKILL.md and the install_mineru function in scripts/mineru_parse.py fetch assets from public registries and model hubs.
[DATA_EXPOSURE]: Documentation describes a configuration file ~/.mineru.json that stores an LLM API key, which could be exposed if the local environment is compromised.
Evidence: The references/api_reference.md file shows a configuration schema containing an api_key field for llm-aided-config.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted PDF files and converts them into structured text for LLM consumption, creating a surface for injection attacks.
Ingestion points: The parse_single_pdf function in scripts/mineru_parse.py reads arbitrary PDF files.
Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between the tool's output and potentially malicious instructions embedded in the PDF.
Capability inventory: The skill environment includes file system write access and the ability to execute shell commands via the wrapper script.
Sanitization: There is no evidence of sanitization or safety filtering applied to the extracted PDF content before it is returned to the agent.

Geek-skills-mineru-pdf-parser