load-docs

Document ingestion loads specified local files, recursively filtered folders, or remote URLs into the agent context completely. This grounds the agent with full content, enabling reliable follow-up question answering and content summarization without manual reading.

Yes, large PDFs are ingested safely using chunk-safe reading that traverses entire page ranges via a checklist. This prevents accidental content truncation, ensuring the full document is loaded into the context before the skill completes.

Yes, document ingestion supports fetching remote URLs via web retrieval and recursively loading local folders with sensible filtering. This allows you to populate the agent context with diverse content sources for knowledge grounding.

The best way to ground an agent is through full-document ingestion. By completely reading specified local files, folders, and remote URLs into the context, the agent can reliably reason from and summarize the ingested knowledge base.

To summarize documents loaded from a folder, first ingest the folder contents recursively into the agent context. Once the full content is loaded and the skill signals DONE, the agent can directly answer questions and summarize the ingested material.

The document loading skill stops after one turn because it operates on a single-turn workflow. It runs once per invocation to completely load the specified documents, then deactivates after signaling DONE to prevent unnecessary context window consumption.