The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill exhibits a significant Indirect Prompt Injection surface (Category 8).
Ingestion points: The script scripts/doc_converter.py ingests untrusted external data from various document formats (.pdf, .docx, .pptx).
Boundary markers: There is no evidence of delimiters or instructions to ignore embedded commands within the processed documents.
Capability inventory: The skill has the capability to read local files, create directories, create symlinks, copy directories to the ~/Desktop, and execute shell commands like open.
Sanitization: No sanitization or validation of document content is documented before it is passed to the Ollama model or used for classification names.
[COMMAND_EXECUTION] (MEDIUM): The skill's workflow involves executing a local Python script with multiple flags that modify the file system and running shell commands (cp -a, open) on user directories.
[EXTERNAL_DOWNLOADS] (LOW): The skill requires external dependencies including markitdown and ollama. While these are from trusted sources (Microsoft and Ollama), they involve installing executable code on the host system.
[DATA_EXFILTRATION] (MEDIUM): The skill utilizes the requests library. While intended for local Ollama API communication, this capability combined with the ingestion of sensitive documents from ~/Documents creates a potential path for data exfiltration if the underlying script is compromised.

doc-mindmap