The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill integrates with several well-known technology companies and cloud services. It uses the LandingAI ADE SDK for document processing, OpenAI for generating embeddings, and Snowflake for database storage. It also utilizes common open-source libraries such as Pandas, PyMuPDF, and FastEmbed.- [COMMAND_EXECUTION]: The documentation provides examples of using Python via shell scripts to perform diagnostic tasks like rendering document pages and inspecting parsed structures. These commands are localized to the user's document processing environment.- [SAFE]: The skill demonstrates security best practices by recommending the use of environment variables for API key management and providing robust error handling in its batch processing examples. No obfuscation, unauthorized data exfiltration, or malicious persistence mechanisms were found.- [PROMPT_INJECTION]: The skill processes untrusted document data (Category 8: Indirect Prompt Injection), creating a potential injection surface. 1. Ingestion points: Document files (PDF, PNG, JPG) are ingested via the LandingAIADE.parse and LandingAIADE.extract methods across all reference workflows. 2. Boundary markers: The skill does not explicitly show the use of prompt delimiters for the extracted text, though it relies on structured JSON schemas which provide a schema-based boundary for the extraction model. 3. Capability inventory: The skill possesses file-write capabilities (saving results to disk) and network operations (API calls to LandingAI, OpenAI, and Snowflake). 4. Sanitization: Basic text cleaning is performed in clean_chunk_text to remove anchor tags from the markdown output. Given the skill's primary purpose is document extraction, this inherent surface does not escalate the safety verdict.

document-workflows