The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/convert_document.py automatically executes shell commands to maintain the environment and run components.
Evidence: The install_dependencies and _ensure_shared_node_modules functions utilize subprocess.run to call pip and npm respectively. Additionally, the convert_md function executes node to process Markdown files.
[PROMPT_INJECTION]: The skill represents an indirect prompt injection surface because it processes untrusted documents and returns the content directly to the agent's context.
Ingestion points: Document content is ingested from local files via scripts/convert_document.py across multiple formats (Word, Excel, PDF, PPTX).
Boundary markers: Absent. The skill returns raw Markdown content without protective delimiters or instructions to the agent to treat the content as untrusted data.
Capability inventory: The agent is intended to summarize, analyze, or respond based on the document's content, allowing instructions embedded in the documents to potentially override agent behavior.
Sanitization: Absent. The logic focuses on text extraction and formatting without filtering for malicious natural language instructions.

bruce-doc-converter