research-doc

Document Research: $ARGUMENTS

Run multi-phase document analysis with progressive memory saving. Retain document/page pointers and check decisive claims against the originals; generated citations and tiers are proposals.

Phase 1: Identify Documents

1. Determine the input from "$ARGUMENTS":

   • If it's a single file path: use research_document with file_paths=["<path>"]
   • If it's multiple paths (comma or space separated): split and pass as file_paths list
   • If it's a directory: use Glob to find all PDFs/text files, then pass as file_paths
   • If it starts with http:// or https://: pass as urls list
   • If arguments include both local files and URLs: pass both file_paths and urls

2. Determine scope from context:

   • Single document, quick question → scope="quick"
   • 1-2 documents, standard analysis → scope="moderate" (default)
   • 2+ documents, comparative analysis → scope="deep"
   • Comprehensive review with methodology critique → scope="comprehensive"

3. Call research_document with:

   • instruction: Use $ARGUMENTS context or default to "Analyze these documents comprehensively — extract key findings, assess methodology, identify agreements and contradictions."
   • scope: As determined above
   • thinking_level: "high"

Phase 2: Present & Save Results

1. Present findings organized by evidence tier:

   • CONFIRMED — Checked against the cited original passage/data; distinguish what the document states from independently established truth
   • STRONG INDICATOR — Strongly implied by document evidence
   • INFERENCE — Reasonable conclusion from document context
   • SPECULATION — Extrapolation beyond what documents support
   • UNKNOWN — Documents do not address this

2. For multi-document results, highlight:

   • Cross-references: Where documents agree or contradict
   • Evidence chains: How findings flow across documents
   • Methodology critique: Per-document reliability assessment