The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from local files (PDF, DOCX, TXT, JSON) and web URLs to build search indices. This creates a surface for indirect prompt injection where instructions embedded in documents could influence the agent's behavior during retrieval.
Ingestion points: Data enters the system through DocumentLoader.loadFile, DocumentLoader.loadDirectory (reading local files), and WebLoader.loadUrl (fetching external web content).
Boundary markers: The skill does not implement delimiters or explicit instructions to the LLM to ignore instructions embedded within the processed text chunks.
Capability inventory: The skill uses fs/promises for local file access, fetch for network requests to arbitrary URLs, and the OpenAI API for embedding generation.
Sanitization: The TextPreprocessor performs text cleaning (whitespace normalization and character filtering), but these are for improving embedding quality rather than security-focused sanitization or instruction filtering.
[EXTERNAL_DOWNLOADS]: The skill relies on several external Node.js packages for specialized document parsing and API management.
Dependencies: Includes pdf-parse, mammoth, jsdom, @mozilla/readability, openai, and p-limit.

embedding-pipeline-builder