The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process external, untrusted document data (PDFs, images) to extract structured text. This creates an indirect prompt injection surface where instructions hidden within documents could influence the agent's logic or subsequent steps.
Ingestion points: Document files ingested via docutray convert, docutray identify, and docutray steps (SKILL.md, references/convert-reference.md).
Boundary markers: Absent. No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within the processed documents.
Capability inventory: The skill performs file system reads, file system writes (via --output), and network requests to the DocuTray API.
Sanitization: Absent. There is no mention of sanitizing or filtering the extracted document content before it is returned to the agent.
[DATA_EXFILTRATION]: The skill transmits document content and metadata to https://app.docutray.com for processing. While this is the intended primary purpose of the skill, it involves sending user-supplied data to an external service. Authentication is managed via the DOCUTRAY_API_KEY environment variable.
[COMMAND_EXECUTION]: The skill makes extensive use of the docutray CLI tool. This involves executing commands on the local system to process files and manage extraction schemas.

docutray-platform